PaperClipAI

0 Fork 0

mirror of https://github.com/paperclipai/paperclip.git synced 2026-11-09 16:02:36 +01:00

Author	SHA1	Message	Date
DottaandClaude Opus 5.5	9f0ebf6f70	fix(accounting): close a run whose turn was lost on restore, and stop succeeded runs holding all work (#19724 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Budgets hold new work while the cost of completed runs is unknown > - In a restart chaos wave, a native ACPX Claude run crossed a server kill, resumed, and succeeded. Its usage receipt never became complete. > - For the 10-minute closure bound, admission refused all work in the company. This included tasks that never crossed a fault. > - This pull request fixes the receipt for a turn that was lost on restore. It also stops a succeeded run from holding all new work. > - The benefit: one late usage report no longer freezes a company, and hard stops still apply. ## Linked Issues or Issue Description Refs #19700 (the accounting closure bound that this follows). What happened? A native ACPX Claude run crossed a server SIGKILL in the middle of a turn. The Runner restored the session, and the run succeeded 54 seconds after it started. All 17 usage receipts stayed incomplete. Accounting settled the run only at the 10-minute closure bound. During those 10 minutes, admission refused all company work with "Company cannot start work while completed runs await accounting". Root cause ACPX reports provisional usage (`runDeltaComplete: false`) until a turn ends. The kill lost the turn that was active, so that turn never got a complete report. The restored Runner failed that turn with `provider_turn_lost_on_restore`. Native turn accounting (`createNativeTurnAccounting`) required every terminal turn to have usage, so this turn kept every later receipt incomplete. This was true even after the continuation turn completed with full usage. The admission gate then counted the succeeded run as a pending terminal run for the whole company. Expected behavior The receipt closes when the run's turns end. A turn that was lost on restore makes the usage a lower bound. A succeeded run with only usage outstanding does not hold other work. ## What Changed - `native-usage-accounting.ts`: a turn that failed with `provider_turn_lost_on_restore` and has no complete usage no longer holds the receipt open. The snapshot marks `accountingUsageLost`. - `native-session-executor.ts` and `native-usage-delivery.ts`: such a receipt carries the provenance note "a provider turn was lost on restore and never reported its usage; observed usage is a lower bound". - `anthropic-pricing.ts` and `codex-pricing.ts`: the rate-card pricers keep a producer's provenance note. - `native-usage-delivery.ts` and `usage-receipts.ts`: remote producer delivery keeps a `lostOnRestore` marker on the receipt's producer event, so replay also excuses the lost turn. Receipts saved without the marker still replay. - `budget-exposure.ts` and `budgets.ts`: admission waits only for pending runs that ended without succeeding (`waitingRunCount`). A succeeded run whose provider reported observed usage is bounded like a running run: its observed cost counts toward spend and reservation capacity, and its settled charge can still reach the hard stop. - `budget-exposure.ts`: a closure that timed out because of a control-plane restart keeps the unpriced-usage exemptions, whatever its token count. Restart evidence is a replaced receipt source, a native recovery transition that is not an ordinary same-run retry, or a turn lost on restore. A second chaos run showed the need. It recovered and succeeded, but its final usage report was lost with the restart, so it settled as unpriced 0 cents. The blocking policy then opened a hard incident and paused the company. A timed-out closure with no restart evidence still counts. - `run-cost-accounting.ts`: a lost receipt that carries a positive provider-reported cost settles at that cost as an estimate marked as a lower bound, not as unpriced. - `paperclip-runner-tool-authority.ts` and `budgets.ts`: a running run's own tool mutations are evaluated as admitted work. Hard stops and pauses still refuse them. Accounting holds, recovering runs and budget inputs that await evaluation do not. In a third chaos run, a live run's write to its task document was refused during a temporary hold, so the run proposed a blocked result. - `doc/SPEC-implementation.md`: documents these rules. Policy choice: a succeeded run's provider work is over, so it has less unknown spend than a running run. Running runs never held admission. Failed, cancelled, timed-out and interrupted runs still hold new work, because their provider may not have stopped. A per-agent gate would still freeze unrelated work of the same agent. A per-reservation gate does not help when `reservationCents` is 0, which was the case in this wave. ## Verification - Live: a restart chaos wave ran master plus this PR (c2a48fb1e0) with no masking. It passed 6/6, and every task crossed a restart. There were 0 accounting holds, 0 refused agent writes, 0 company pauses, and every cost settled as estimated. - Red then green: `native-usage-accounting.test.ts`, "closes a run whose earlier turn was lost on restore, as a lower bound". It uses the event sequence of the chaos run. - Red then green: `closed-native-receipt-accounting.test.ts`, "does not hold other work for a succeeded run with only usage outstanding". It also checks that the hard stop still applies. - I changed "neither pauses nor holds admission while a succeeded run's charge commits" to the new policy. A failed run in the same state still holds new work. - Red then green: `closed-native-receipt-accounting.test.ts`, "never hard-stops a blocking policy for a restart-derived timed-out receipt". It uses the zero-token row shape of the chaos run and a case with unpriced tokens. Without the fix, the company was paused. - Red then green: `runner-api.integration.test.ts`, "lets a running run write during a temporary accounting hold, but not after a hard stop". - Red then green: `closed-native-receipt-accounting.test.ts`, "does not hold other work for the s7504 succeeded row shape". - Red then green: `closed-native-receipt-accounting.test.ts`, "settles a timed-out restart receipt at its provider-reported cost, without holding or pausing". It uses the OpenCode/OpenRouter row shape from a fourth chaos run. - `anthropic-pricing.test.ts`: the rate-card provenance keeps the lower-bound note. - Local: `closed-native-receipt-accounting`, the budget suites, `native-usage-accounting`, `native-usage-delivery.integration`, `native-runner-restart-recovery.integration` and `anthropic-pricing` all pass (187 tests). Server `tsc --noEmit` is clean. `pnpm verify` asks for CI coverage because the spec changed. ## Risks - A succeeded run whose final charge is higher than its observed cost can let one more task start before the charge commits. The hard stop then pauses the scope as it does for a running run that overruns. - A run that includes a lost turn settles with a lower-bound usage figure. The provenance says so, and a later receipt still reconciles through the late-receipt path. ## Model Used - Anthropic Claude Opus 5.5 (`claude-opus-5-5`), Claude Code agent with tool use and code execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 23:57:48 -06:00
Devin Foley Sentry Bot Claude Opus 5.5 Paperclip	131b566ca4	fix(workspaces): report safe Git restore diagnostics (#19736 ) Return the restore stage and failure category with the existing HTTP409 rejection. Preserve the unsaved comment and editable draft. Keep nested saved-commit and clone-timeout diagnostics through preparation wrappers, and distinguish local Git file errors from authentication failures. Retain only safe process flags privately; no raw command cause is exposed. All seven affected local suites pass (288 tests) and the layered server typecheck passes on supported Node25.6.1. Reviewed head `022eb7c2eb` has Greptile Apex5/5, zero unresolved findings and all applicable CI passing (55 successful checks and4 intentional skips). Co-Authored-By: Sentry Bot <sentry-bot@paperclip.invalid> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-11-06 21:33:59 -08:00
Dotta Paperclip Codex	839d41aabb	fix(server): retire owned plugin listeners before idle sleep (#19358 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Plugins receive durable agent lifecycle events through a registered host. > - The host uses a PostgreSQL LISTEN connection to wake delivery. > - An idle report could return `none` while that host-owned connection stayed open. > - This change gives that listener an exclusive owner and closes it during the current idle hold. > - Release or expiry restores listening before journal replay. > - Failed close and pending work continue to block sleep. ## Linked Issues or Issue Description Refs #18988, which is now merged. This focused ten-file change now targets master. Current integrated candidate: `0f83005122ef1bdb1f1e08f8f58424c9a840ebb1`, with actual master `644ee66f5d4ae8365fe8b9eca57df2be7405b5ab` preserved through normal merges. No history was rewritten. What happened? A normal local App with an installed lifecycle plugin could return an idle `none` report while its lifecycle LISTEN backend and socket stayed open. Expected behavior The host must close its own listener before it reports idle safety. It must retain other subscriptions, durable events, company scope, retry deadlines and useful reactivation. Steps to reproduce Start the normal server with a lifecycle plugin. Create an agent and wait for its durable acknowledgment. Stop the server, create another agent while it is stopped, and restart with the same files. Request an owned idle hold after ordinary replay. Check the exact listener backend and matched Node socket before shutdown. Let the valid hold expire naturally, then make an ordinary authorized agent update and check its durable plugin acknowledgment. Live executing or persistent preview services remain active. The separate hosted preview dependency [paperclip-cloud#478](https://github.com/paperclipai/paperclip-cloud/pull/478) retains its own App and operator prerequisites. This PR does not claim deployed preview or suspension acceptance. ## What Changed - Add a private, exclusive database listener owner that inherits the normal client policy. - Prepare plugin lifecycle retirement after native retirement under the same current idle hold. - Use one delivery-eligibility helper for dispatch and idle journal inspection. It preserves the same manifest, capability, company-disable query, and skip rules. - Close only the owned listener. Keep failed close ownership and inspection failures visible. - Resume listening on normal release or expiry, before journal replay. - Preserve disabled-feature retry history and the original deadline on re-enable. - Await lifecycle disposal during normal shutdown and consumer teardown. - Add owner regressions and document the lifecycle contract. ## Verification Current integrated source: `0f83005122ef1bdb1f1e08f8f58424c9a840ebb1`. Fresh normal Runner TypeScript passed in 0.785 seconds. The complete normal App layered semantic check passed in 28.284 seconds. All tracked source, the lockfile and the 18 separate private witness inputs stayed unchanged. Known compiler owners closed and the isolated temporary home was removed. Current product qualification: at measured integration `896cc9b62993e6ed10418d7a7417e1ddbf8dc89c`, all three whole files passed with no failures or skips: lifecycle delivery 40, runtime hooks 20, and runtime assignments 24. The actual physical LISTEN and natural-expiry controls passed. All owned clients and captured processes closed. Independent final session checks were zero before normal deletion of each successful test database. Normal full-App physical retirement: the original case passed in 47.574 seconds. It exercised real startup, readiness, SDK callbacks, durable ACK, foreign-company exclusion, physical stop, offline input and cold replay. The initial unknown hold and its ordinary release remain in the evidence. After the original 30.5-second cadence, a new valid hold returned quiescent. The exact current cold-App LISTEN backend and its matched Node socket closed before server stop. Normal full-App natural expiry: the separate original case passed in 77.391 seconds under its original 90-second body cap. The final valid 30-second hold expired naturally. A new listener backend and matched current App socket appeared before fresh input. An ordinary authorized agent-name PATCH produced journal event 20, the actual SDK callback and durable ACK. Foreign exclusion and readiness assertions passed. The final measured hold had no manual release, replacement hold or shortened TTL. The original preparatory unknown hold, ordinary release and 30.5-second cadence remain explicit. Both App cases used the normal current server helpers and actual SDK bundling/worker entry. A private compatibility change supplied the same owned HOME at the exact existing child-fork boundary. It did not change the product helpers, business assertions or timing limits. These separate acceptance fixtures remain outside the focused product diff. Source carry to the current candidate: all nine other owned files, listener mechanics, actual producer paths, SDK paths and measured hold behavior are unchanged. The sole owned intersection from newer master is an independent environment-delete cleanup callback in `app.ts`. The plugin witnesses do not call that route. Normal merges preserve every unrelated master blob and file mode. The fresh current semantic checks above qualify the integrated source. New master Native crash-recovery and saved-copyback Retry capabilities are not claimed by these plugin cases. Actual master merge target: newer master contains retained plugin Runner release/resume and other Native recovery fixes. The projected combined target passed fresh normal Runner TypeScript in 0.799 seconds and the full App semantic check in 29.715 seconds. All ten plugin-owned blobs and modes remain exact to the published head; every unrelated master change is preserved. The measured plugin cases create no Native provider turn or warm Native session. The unchanged native-first/plugin-second ordering still checks the native result before listener retirement. Later sparse-checkpoint and Native response-publication changes do not enter those paths. This is a scoped source carry, not new hosted Runner acceptance. Failures retained: an earlier reactivation fixture's real timer returned one millisecond before its original wall-clock expiry. Its private fixture now waits any positive remainder and retains the same expiry assertion and limits. The earlier partial-composition `unknown` result remains distinct from normal-startup proof. Current qualification also exposed missing normal workspace dependencies and a private guard that assumed loose Git refs. Normal owned dependency setup and an exact loose-or-packed reader fixed those setup failures. No business assertion, timeout, permission boundary or product sleep gate was relaxed. Repeat focused product coverage with a fresh external fixture database and the normal server Vitest command, one worker: ```sh pnpm --filter @paperclipai/server exec vitest run src/__tests__/plugin-lifecycle-delivery.test.ts --maxWorkers=1 --no-file-parallelism pnpm --filter @paperclipai/server exec vitest run src/__tests__/agent-runtime-hooks.test.ts --maxWorkers=1 --no-file-parallelism pnpm --filter @paperclipai/server exec vitest run src/__tests__/agent-runtime-assignments.test.ts --maxWorkers=1 --no-file-parallelism ``` Exact current head `0f83005122ef1bdb1f1e08f8f58424c9a840ebb1` has 60 successful checks, three skips and one neutral result, with no pending or failing contexts. Greptile completed its full review at 5/5 with no actionable issue. All root and nested review pages were read; the earlier eligibility finding is resolved. Ponytail's latest turn stopped before a verdict. A normal retry was requested; no current Ponytail verdict or score is claimed. Previous `3ca86ec` CI and scores are historical. A full local repository typecheck/test/build or browser run is not claimed. Controlled local App/SDK/PostgreSQL proof does not establish live provider, deployed Railway/Neon suspension, signed Cloud alarms or global exactly-once delivery. ## Risks - A listener close failure or uncertain inspection must continue to block sleep. - Release, expiry, configuration changes, and journal arrival can race with retirement. Epoch and current-hold checks retain conservative results. - Delivery remains at least once across uncertain ACK boundaries. This change does not add global exactly-once delivery. - The database helper closes only its exclusive listener client. Other owners and subscriptions keep their resources. ## Model Used OpenAI `gpt-6.1-sol`, reasoning and code/tool use. The managing Codex agent implemented shared database/composition seams and reviewed the change; its exact model ID was not exposed. The runtime does not expose a verified context-window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing> Co-authored-by: Codex <noreply@openai.com>	2026-11-06 22:56:47 -06:00
DottaandPaperclip	1c91bacf97	fix(native): admit independent replies during contended publication (#19726 ) ## Thinking Path > - Paperclip manages work done by AI agents. > - A completed native run must deliver its saved final response. > - The response worker also recovers retained replies that can wait on publication locks. > - A pass waited for its old publication groups even when other check slots were free. > - This could delay an independent final reply beyond a full page of retained candidates. > - This PR enables the existing shared coordinator option that admits independent due work during an active pass. > - Existing authority checks, resource keys and the four-check limit still control publication. ## Linked Issues or Issue Description What happened? A native run completed after replacement recovery. Its final response and accounting were saved, but its final issue reply was missing. The response queue contained more than 100 retained chat replies and repeated endpoint-lock failures. The existing worker did not admit independent due entries until its current groups finished. Expected behavior A contended publication owner must not hold free worker slots. An independent accepted final reply must publish once, including when its candidate is on a later SQL page. Steps to reproduce Create 100 accepted, committed replies that share one publication owner. Create an independent reply after them in candidate order. Hold the old coordinator rows with a database transaction and start the real response worker. Before this change, the independent reply stays unpublished until the old group is released. Related: #18977 changes response selection in standing conversations. This PR changes only worker admission and preserves response selection. ## What Changed - Enable `admitDueWhileActive` in the native response worker's existing shared coordinator configuration. - Add a real PostgreSQL regression with the full 100-row SQL page, the default dispatch batch, real row-lock contention and a later-page independent reply. - Verify one comment per reply after settlement and no further SQL or scheduled deadline across 60 seconds of idle time. - Update one existing principal/rescope integration assertion to observe the reconciled invalidation revision. Keep its actual per-run replies, authorization, publication and idle checks. ## Verification - Regression on the unchanged worker: the independent reply remained unpublished while the first 100 replies were locked. The real PostgreSQL test failed as expected. - Final head `748fb9609104c016c33ff7caf282ba9e15e88495`: all six publication regressions and both principal/rescope controls passed (8/8, 47.96 seconds). Checks cover later-page admission, exact per-run comment counts, retry and rollback behavior, authorization, and 60 seconds with no idle SQL or scheduled deadline. - The tests used the canonical PostgreSQL wrapper and one owned disposable PostgreSQL 18.1 fixture. All clients and test databases were closed before fixture cleanup. - The current-head [full CI run](https://github.com/paperclipai/paperclip/actions/runs/42338953241) supplies typecheck, build and broad test coverage. All current-head checks passed; the merge gate is green. - Local full typecheck was blocked by unbuilt dependencies; it is not counted as a pass. The prior timer-state CI failure and the final successful rescope-generation regression remain recorded. - Greptile reviewed the final head at 5/5 with no unresolved threads. `git diff --check` passed. - Deployed acceptance remains separate. No serving instance was changed by the PR tests. ## Risks The worker can now use free slots during a slow pass. The shared coordinator still limits checks to four and preserves task, endpoint and principal exclusions. Authorization, result identity, retry limits, response selection and transaction ownership are unchanged. The regression checks idle quiet after all concrete obligations settle. ## Model Used OpenAI GPT-6.1 (`gpt-6.1-sol`), with reasoning, code execution and tool use. The runtime does not expose the context window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 22:47:30 -06:00
DottaandClaude Opus 5.5	aed053633e	fix(environments): run pending sandbox cleanup when an environment is deleted (#19716 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Operators delete a sandbox environment, optionally with consent to destroy its sandboxes (`DELETE /api/environments/:id?destroyReusableSandboxLeases=true`) > - A sandbox whose teardown failed or was interrupted is kept as a `pending_cleanup` lease, and it blocks the delete > - The cleanup sweep retries such a lease only after a cooldown of up to 5 minutes > - A consented delete therefore returned 409 `pending_sandbox_cleanup` for minutes. A caller with a shorter deadline gave up, and the environment and its sandbox stayed behind > - This pull request makes the consented delete run one teardown attempt for the environment's pending cleanups now, then re-check > - The benefit is that a consented environment delete completes as soon as the provider teardown succeeds ## Linked Issues or Issue Description What happened? After a restart chaos wave, the harness deleted the wave's Daytona environment with `destroyReusableSandboxLeases=true`. The server returned 409 `pending_sandbox_cleanup` every 3 seconds for the whole 120-second client deadline. A run lease had entered `pending_cleanup` after its run failed, and the sweep's retry cooldown had not passed. The environment was not deleted. Expected behavior With explicit consent, the delete tears down the environment's own pending sandbox cleanups and completes when that teardown succeeds. A teardown that fails, or one already in flight elsewhere, still blocks the delete. Steps to reproduce 1. Leave a sandbox lease of an environment in `pending_cleanup` with a future retry time. 2. `DELETE /api/environments/:id?destroyReusableSandboxLeases=true`. 3. On master, the response is 409 `pending_sandbox_cleanup` until the cooldown ends. Refs #19696 (restart recovery for creation and teardown claims; this builds on it). ## What Changed - `heartbeat/recovery.ts`: `sweepPendingCleanupLeases` accepts `environmentRetry: { environmentId, actorId }`. Like the existing run-scoped explicit retry, it bypasses the cooldown for one attempt, never takes over an in-flight attempt, and logs `environment_lease.cleanup_retried` with reason `environment_deleted`. It touches only that environment's leases. - `routes/environments.ts`: when the delete has consent and the only blockers are pending cleanup and reusable leases, it runs that retry first, then re-reads the blast radius. The existing reusable-lease destroy and the final guard are unchanged. The new dependency is a narrow, optional `retryPendingEnvironmentCleanup` option; without it, behavior is unchanged. - `app.ts`: wires the option to the existing heartbeat service instance. ## Verification - Red, then green: - `environment-routes.test.ts`: "runs the environment's pending sandbox cleanup now and deletes with explicit consent" and "keeps rejecting when the consented pending cleanup attempt does not finish" fail before and pass after. "does not run pending sandbox cleanup without consent or past another blocker" passes in both. - `heartbeat-pending-cleanup-sweep.test.ts`: "runs one attempt for a deleted environment's cleanup during its cooldown, scoped to that environment" fails before and passes after. It checks that another environment's lease and an in-flight claim are untouched, and that the activity log is written. - `sandbox-creation-operations.test.ts`: "reclaims an issue-bound creation after its run failed and the issue finished on a later run" covers another chaos finding. A creation intent bound to an issue was interrupted by SIGKILL 0.5 s after it committed, its run ended `process_lost`, and a later run finished the issue. The test fails on the code before #19696 and passes on master, so #19696 already covers this case. The test is kept as a regression. - After rebasing on master, `environment-routes`, `heartbeat-pending-cleanup-sweep`, `agent-runner-qualification-routes` and `sandbox-creation-operations` pass (192 tests). `tsc --noEmit` reports no errors in the changed files. ## Risks - A consented delete now makes provider teardown calls for pending cleanups. Those sandboxes were already queued for destruction, and the operator consented to destroy the environment's sandboxes. - One explicit attempt counts toward the lease's attempt cap, the same as a user Retry. - The delete request now waits for those teardown attempts. Each attempt uses the existing plugin timeouts. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file editing, test execution). Extended reasoning enabled. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 21:28:54 -06:00
DottaandClaude Opus 5.5	51eb4ff705	fix(recovery): finalize an accepted native result after a hard kill (#19713 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - A native run hands its accepted result to a controller lease, and the controller finalizes the task from that result > - A graceful shutdown releases that lease for the next server (#19591), but a hard kill (SIGKILL, OOM, host crash) cannot > - After a hard kill between result acceptance and finalization, the next server waited for the dead controller's 20-minute lease, so the run stayed `running` and the user's follow-up comment waited with it > - This pull request releases that lease at startup when the controller is proved dead by its boot ID and process identity, never by age > - The benefit is that the saved result finalizes within seconds after a restart, without another provider turn, and queued follow-up work continues ## Linked Issues or Issue Description Refs #19591 (the graceful-shutdown counterpart) and #19578 (boot-identity proof for legacy runs). What happened? A restart-chaos wave sent K1 (SIGKILL) 20 ms after a Claude run recorded `run.result.accepted` and `run.terminal`. The run then stayed `running` for more than 10 minutes, through four more restarts. The finalization row was `workspace_finalizing` with the stored result. Its lease belonged to the killed server: a different `controller_boot_id`, a `controller_pid` with no process, and `lease_expires_at` 20 minutes ahead. Startup recovery skips runs that already have a result. The result-commit worker waits until the lease expires. The task's follow-up comment could not run until then. Steps to reproduce 1. Let a native local run accept its result. 2. SIGKILL the server before finalization finishes. 3. Restart the server. The run stays `running` until the old lease expires. Expected behavior After a restart, a result lease held by a stopped controller of an earlier boot is treated as released at once. The stored result finalizes once, with no new provider turn, and the follow-up wake runs after it. ## What Changed - New `server/src/services/native-runtime/native-dead-controller-result-lease.ts`: `releaseDeadControllerResultLeases` finds `running` native runs with an accepted result and an unexpired lease. It requires the lease to belong to a different controller boot. It proves the controller stopped with `evaluateNativeControllerTakeover` (`requireStoppedController`): the process is dead, or its PID now names another process. It then expires that lease with a compare-and-set on the exact owner token, boot ID, and PID, and notifies the result-commit and finalization workers. - `server/src/services/heartbeat/recovery.ts`: `recoverNativeRunsAfterRestart` calls this first and records a `native.recovery.transition` event (`awaiting_result_finalization`) for each released run. - Only executor leases are released, because only the executor claim writes the controller identity together with the lease. Finalizer (`native-finalizer:`) and cleanup (`native-cleanup:`) claims keep an earlier executor's identity, so they are never released this way and expire on their own short leases. A finalization-resume token (`:resume:`) on a result row always belongs to the executor that adopted it, because a pending resume claim is taken only on a row without a result. - Also not released: a live owner, an owner without a recorded process, a lease of the current boot, a run this process is executing, and remote workspaces (`nativeWorkspaceSync` or `remoteWorkspace`), which keep their fenced lease. - The compare-and-set does not compare the stored expiry. Postgres stores microseconds that a JavaScript `Date` drops, and an exact-expiry comparison silently skipped the real lease in the live reproduction. - `doc/execution-semantics.md`: documents the rule. ## Verification - Red then green on embedded Postgres (`server/src/__tests__/heartbeat-native-restart-suspension.test.ts`): - A hard kill with a dead controller's executor lease (first claim, restart claim, or resumed claim) now finalizes on the next startup: run `succeeded`, finalization `committed`, task `done`, one result, one run, no provider dispatch. These tests failed before the fix with `status: running`. - A lease expiry stored with microsecond precision is released. This test failed against the first version of the fix and reproduces the live miss. - A live controller of another boot keeps its lease, and so does a lease of the current boot. - A live claimant's finalizer or cleanup lease is kept when the recorded earlier executor is dead. - `npx vitest run src/__tests__/heartbeat-native-restart-suspension.test.ts`: 20 passed. - Related suites pass: `native-session-resumption`, `native-runner-restart-recovery.integration`, `native-run-projection-work`, `live-recovery-work`. One `live-recovery-work` legacy case timed out once under load and passed when run again by itself. - Live, local: `tests/restart-chaos/main.ts --workloads C1 --triggers result --faults K1 --rc06-masked`. The failing case above came from such a wave. I started its kept database and confirmed the stored lease state and the compare-and-set miss. Against that same database, the fixed release expires the stuck executor lease. Two later waves (12 tasks, `--triggers result`) passed every task. Run events were not kept for passing tasks, so I cannot confirm that a kill landed in the narrow pre-finalization window in those waves. - Live proof on head `8971874654`: local ACPX Claude, workload A1, seed 7411, `--triggers result` (a local variant that fires on A1 2.5–3.5 s after the run's result proposal), `--keep`. The run passed 4 of 4 tasks. In two tasks, K1 landed 1.5 s and 2.0 s after `run.result.accepted`. On the next startup, each run recorded "Server restart released the accepted native result of a stopped controller for finalization" (`controller_process_dead`). Each finished `succeeded` about 1 s later, with its task `done` and exactly one run per task (no extra provider turn). Before this fix, the same shape stayed `running` until the 10-minute budget ran out (seeds 7411, 7422, and 7423 from the RC-33b investigation in #19691). ## Risks - Startup could take a lease from a finalizer that is still running. Mitigations: the lease must belong to another boot, the controller process must be proved stopped on this host (the same proof the existing runner takeover uses), and the update matches the exact owner. Remote workspaces are excluded. Multi-host deployments keep the existing PID-liveness assumption that `claimNativeRestartRecoveries` already makes. - Low behavioral risk otherwise: finalization uses the existing reconciler path, which #19591 already exercises after a graceful drain. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, run as an agent in Claude Code with tool use (shell, file editing, local test execution, live restart-chaos waves). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 21:22:24 -06:00
DottaandClaude Opus 5.5	08f7ac03dc	fix(recovery): keep a never-started task waiting through a temporary accounting hold (#19702 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Admission checks every wake. A short company accounting hold can refuse new work after a control-plane restart. > - When the hold refused a new task's assignment wake, stranded-work recovery found the task with no run while the hold was still active. It moved the task to `blocked` with a board recovery comment. > - A `blocked` task leaves the stranded-assignment sweep, so nothing woke it after the hold cleared. The task stayed `blocked` with no run, no interruption and no Retry. > - The hold was temporary. Tasks created a few seconds later (after the sweep) started normally once accounting settled. > - This pull request keeps a never-started task `todo` while a temporary budget hold lasts. The existing assigned-todo liveness dispatch then admits it once the hold clears. The task shows the hold it waits for. > - The benefit is that a restart no longer silently drops new tasks created during the accounting settle window. ## Linked Issues or Issue Description What happened? In a restart-chaos storm on master `06bbcea12f`, three new tasks were created while the company accounting hold was active. Each assignment wake got 409 "Company cannot start work while completed runs await accounting." (or "...native run recovers its unfinished accounting."). One task rested `blocked` with 0 runs, interruption `null` and `canRetry=false`. The other two ran later. Root cause `reconcileStrandedAssignedIssues` (`server/src/services/recovery/service.ts`, the `readInvocationBudgetBlock` branch near line 4826) escalates any assigned task to `blocked` when a budget block exists. It does this before the assigned-todo no-run branch, which already skips a budget-blocked task and dispatches it later. So a task that never started was treated as failed recovery work. The refused wake itself is stored as a `skipped` `budget.blocked` diagnostic, which is correct; the escalation removed the only path that re-admits it. Expected behavior A wake refused by a temporary admission hold waits and is admitted once the hold clears, with one run. A real hard stop is still a board decision. Removed assignees and cancelled tasks still do not wake. Steps to reproduce Create a company with a hard-stop budget and a completed run that has `cost_accounting_pending = true`. Assign a new `todo` task. The wake gets 409. Run the stranded-assignment sweep. Before this change the task becomes `blocked` and never runs. Related: #19700 bounds the accounting hold itself. This PR is independent: it handles what happens to a task whose first wake the hold refused. ## What Changed - Stranded-work recovery skips a `todo` task with no run while a temporary budget hold blocks admission. It does not escalate it to `blocked`. The existing assigned-todo liveness dispatch admits the task once the hold clears. - A hard stop (spend at the limit, or blocked unpriced usage) keeps the existing board escalation. Recovery checks this with the read-only `policyBlocks` for the policies of the blocking scope. It does not change budget hold semantics. - The task interruption projection now covers a `todo` task with no run whose latest wake was refused with `budget.blocked`. It shows "Waiting for the budget to allow new work. The task starts automatically when it does." with no Retry. A budget pause shows the existing "budget" user action. - The no-run projection applies the same person-owned gates as the run-backed projection: a budget pause shows the "budget" action, and an ordinary agent, project, company or task pause shows "resume". Only an accounting hold shows the automatic-start text. - `ExecutionBlockerNotice` shows that wait text for a `working` presentation that carries a wait reason. - `doc/execution-semantics.md` records the rule. ## Verification - New embedded-Postgres regression `server/src/__tests__/heartbeat-refused-assignment-wake.test.ts`: - hold active, new task assigned, wake refused (409, `skipped` `budget.blocked`), two recovery sweeps: the task stays `todo` with no run and shows the wait reason. Before the fix this failed with `expected 'blocked' to be 'todo'` (the CHA-9 shape). - hold cleared, two sweeps: exactly one provider turn starts, from `issue.assigned_todo_liveness_dispatch`, and the task finishes. A later sweep does not start another turn. - counter-tests: a terminated assignee and a cancelled task get no run after the hold clears. - Existing `creates a board recovery action for budget-blocked assigned work` (real hard stop) still passes unchanged. - `pnpm exec vitest run src/__tests__/heartbeat-refused-assignment-wake.test.ts src/__tests__/heartbeat-process-recovery.test.ts src/services/task-interruption-presentation.test.ts src/__tests__/heartbeat-issue-liveness-escalation.test.ts` (server) and `src/components/ExecutionBlockerNotice.test.tsx` (ui). - Server typecheck and `pnpm check:token-gates`. - Live restart-chaos storm (local, codex/claude/opencode, 6 in flight, K2 every 3-4 minutes for 20 minutes, no RC-06 mask), revision `e38bb5c8f7`: 17 tasks, 10 pass. - Two new tasks (CHA-16, CHA-17) were created while the company accounting hold was active. Their assignment wakes got 409 "Company cannot start work while completed runs await accounting." Both stayed `todo` (not `blocked`) and showed `state: working`, `canRetry: false`, `retryWait.reason: budget.blocked`, "Waiting for the budget to allow new work. The task starts automatically when it does." No task rested `blocked` with 0 runs. - The hold itself never cleared in this wave (`pendingRunCount: 1` at wave end), so these two tasks did not start before their deadline. That stuck hold is the separate accounting-closure defect that #19700 bounds. The other 5 failures (`R3:retry_no_new_run:B5`) are Retries refused by the same stuck hold. The re-admission after the hold clears is proved by the embedded-Postgres regression above. ## Risks - Low. The change narrows one escalation to tasks that never started, under a budget block that is not a hard stop. Those tasks now stay `todo` and are re-checked by the existing 30-second stranded-assignment sweep, which already re-checks every admission gate. - The hard-stop check reads the blocking scope's active policies once per budget-blocked never-started task in a sweep. It runs only while a hold is active. - The interruption projection adds one indexed read (company + payload issue id) for `todo` tasks in the batch. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file edits, test execution). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [ ] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 21:12:00 -06:00
DottaandClaude Opus 5.5	8a34c1b62c	fix(budgets): bound the accounting hold so one missing receipt cannot wedge a company (#19700 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Pending run accounting holds new company work until it settles (#19497, #19606, #19673) > - A run settles only with closure evidence: a verified local stop, or a terminated remote sandbox. Without that evidence it held the whole company forever. In restart-chaos storm 4, a single run held 35 tasks for hours > - Two more shapes kept the hold alive. A run whose result had already been accepted never counted as closed. A recovery that already owned its held reservation was refused by other tasks' recovering accounting, so concurrent recoveries waited on each other > - This pull request bounds the hold, closes accepted-result runs on a verified stop, and lets an owned recovery finish > - The benefit is that no single missing receipt can wedge a company, while genuinely unknown spend still surfaces under the blocking policy ## Linked Issues or Issue Description Follow-up to #19497, #19606 and #19673. What happened? - Storm 4: run `f879895f` timed out in `session.open`. Its local stop was never verified (its process identity is still recorded), recovery ended in `terminal_failure`, and its receipt never became ready. From then on, every new-work wake got 409 "New work must wait while a native run recovers its unfinished accounting". That was 93 refusals, 35 blocked tasks, and no Retry. - Storm 4: run `6d3c76bf` was resuming as the recovery owner with its own held reservation. `reserveRunBudget` refused it because other tasks' runs were recovering. - Wave s7401: run `0a3a849d` failed after its result was accepted, with a verified local stop. It had 10 receipts across 2 sources, all incomplete with 0 tokens. It stayed "completed runs await accounting" because closure only covered coordinators in terminal failure with no result. What did you expect? A run without closure evidence holds new work only for a bounded time, then settles visibly. An accepted-result run with a verified stop closes like a terminal failure. A recovery that already owns admitted work can finish. Steps to reproduce 1. Restart the control plane during several native turns so that one run's local stop is never verified. 2. Observe the run stay "recovering" and every new wake refused indefinitely. ## What Changed - `run-cost-accounting.ts`: - A coordinator counts as settled when it ended in terminal failure without a result, or when it has an accepted result. The verified-stop and remote-termination closures use that rule, which covers the s7401 shape. - A native run whose provider never started settles at once, like a pre-provider failure, with no charge and its reservation released. Its recovery has settled, no usage receipt was ever journaled (a provider turn journals one as soon as it starts), and its local process never launched or was verified stopped (wave s7403, run 3c1d7b70, `session.open` timeout). - New bound `ACCOUNTING_CLOSURE_TIMEOUT_MS`: the remote runner reconnect grace (7 minutes) plus 3 minutes. A terminal native run with a settled coordinator and no closure evidence settles at the bound: - its known usage is recorded as unpriced (or estimated from the rate card), with provenance evidence "closure evidence never arrived…"; - it is marked `accountingClosureTimedOut`; - a `cost.accounting_closure_timed_out` activity entry names the run. - The reconciler selects these runs, and `readCostAccountingWork` schedules a wake at the bound. - `budget-exposure.ts`: a timed-out closure's final charge is exempt from none of the unpriced-block exemptions. Under the `block` policy it raises the normal budget notice ("recorded usage has no reliable price") instead of a silent hold. - `budgets.ts`, `budget-reservations.ts`, `run-dispatch/adapters/postgres.ts`: when a validated recovery owner continues admitted work, admission and both dispatch rechecks apply only hard stops and pauses. Pending and recovering accounting holds still gate new work. - `doc/SPEC-implementation.md`: documents the bound, the accepted-result closure and the provenance. ## Verification - `closed-native-receipt-accounting.test.ts` (embedded Postgres). All of these fail on master and pass here: - s7401 shape (accepted result, verified stop, all receipts incomplete with 0 tokens): settles and clears admission. - A run without closure evidence: held before the bound with a scheduled wake, then settled with the "closure evidence never arrived" provenance, the `accountingClosureTimedOut` flag and an activity entry. Admission clears under `allow`. - Storm 4 `f879895f` shape (terminal failure, recorded process identity, no stop event): held, then released after the bound. - The blocking policy still blocks on a timed-out closure. - s7403 shape (no receipts, model unknown): a run whose process never launched, or was verified stopped, settles at once with no charge and its reservation released. With an unverified process it waits for the bound. - Storm 4 `6d3c76bf` shape: a recovery owner with its held reservation reserves past another task's recovering accounting, while new work for another task still waits. - Existing suites pass locally: budget-enforcement-scoping, budget-enforcement-worker, native-stop-accounting, cost-data-quality, heartbeat-cost-accounting, run-cost-accounting, budgets-service. Server typecheck passes. - The other storm 4 terminal-failure runs (`2e477ae9`, `e4d2e14d`, `6d3c76bf`, `7d21a44c`, `4bc49df0`, `1d85fbd9`) were already settled by #19673 in that database (`cost_accounting_pending=false`) and were not holding admission. `f879895f` alone held it. ## Risks - After the bound, a run whose provider might still be running settles as unpriced. Under the `block` policy that raises a hard-stop budget notice, by design. Under `allow`, work continues and a late receipt still reconciles. - A recovery owner skips accounting holds, so concurrent recoveries can spend while other tasks' accounting is unknown. Hard stops still apply, consistent with best-effort budget enforcement. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file editing, embedded Postgres tests). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 21:02:01 -06:00
Dotta Paperclip Codex	18be7ac696	fix(server): retire idle native sessions before tenant sleep (#18988 ) ## Thinking Path > - Paperclip lets people manage AI agents and their work. > - Native agents can retain a warm session between useful turns. > - That owner must close and save its files before the tenant can sleep. > - A failed close must keep the owner visible until retirement succeeds. > - This PR prepares idle resources under the exact current admission hold, then takes a fresh safety snapshot. > - Busy turns, uncertain closure and changed holds still prevent sleep. > - CI also exposed a chat authorization race. The fix repeats full authorization after the exact delivery commits. ## Linked Issues or Issue Description Refs #18639 for local work and incoming wake boundaries. Related #18829 covers suspension receipts after admission. What happened? Idle preflight did not retire warm Native owners. Three failed-close paths could forget an owner without proving that its provider stopped. CI also found a false chat rejection when a delivery changed from processing to processed between two validation reads. Expected behavior Retire only idle owners under the exact bounded hold. Wait for closure and file collection. Retain failed owners. Scan safety after preparation. Keep diagnostic reads passive. A newly committed chat delivery must pass the full current authorization checks before attestation succeeds. Steps to reproduce Finish a warm Native turn. Start an idle drain and request safety with its owner ID. Delay or reject provider closure, or replace the hold during preparation or scanning. The report must keep the instance awake. For the chat race, commit the exact delivery between the first binding read and its fallback read. Paperclip version or commit Current head `7f6f02b611dcf5a9a2b05173a44e9631eaf78058` normally integrates actual master `1aa0c3639714f4976cfa56f95aae6e2b7d1235ac`. The merge preserves every unrelated master path. Fresh exact-head CI and complete review remain required. ## What Changed - Retain uncertain Native owners after failed restart, idle timeout and failed-turn closure. - Prepare Native retirement through the existing heartbeat facade. Reuse provider closure and stopped-file collection. - Track preparation under the exact idle owner and expiry. Reject new activity, changed holds and expired holds before and after awaited work. - Take the final durable and local safety snapshot after preparation. Keep diagnostic inspection passive. - Cover ephemeral workspace retirement and useful next use, supported plugin drain and SDK persistence, busy owners, failed closure and hold replacement. - Keep dedicated recovery storage explicit. Use the same validated options for support and storage. Probe-free helpers retain the original implementation bodies and preserve default callers. - Complete the Native fixture through real retirement and budget evaluation. Keep its rejection, file, credential, accounting and strict NONE assertions. - Repeat the complete chat authorization once if the fallback sees the exact batch now processed. Preserve owner locks, principal and binding checks, denial handling and the original retry deadline. ## Verification - The support-mode regression reproduces one failure with four passes under the old policy, then passes all five controls. It retains prior allocation and cleanup cases. The explicit mode does not probe embedded PostgreSQL or use an admin alias. - The deterministic chat race baseline reproduces two authorized failures with all four current-authority denial controls passing. Both Slack and Discord use real PostgreSQL and the real authorization function. The negative setup commits before the attestation transaction; it does not claim to revoke access through a row lock. - The product fix passes normal own App declaration emission with no public declaration changes, plus both current normal owning compiler programs (399 and 611 roots). The same six controls pass after the fix. The full original file plus these controls passes all 30 tests with no skips. - The original controlled Native continuation passes: one test, no failures or skips, 1.718 seconds in its body and 23.138 seconds in its launcher. The original 40-second case and 180-second outer bounds are unchanged. It verifies held closure, strict NONE, an ordinary authorized comment, saved session/files, useful output, encrypted credentials, accepted results and settled accounting. All ten observed processes and groups close; worker connections and listeners reach zero. A separate post-owner audit finds zero sessions; the successful owned database is normally removed. - Earlier preparation, HTTP, lifecycle, workspace and plugin controls remain source-specific evidence. They are not a claim that every final-tree test ran locally. - The original CI assertion log is retained. The controlled race proves a possible cause; the original CI SQL schedule was not captured. Two other jobs lost their self-hosted runners and provided no test verdict. Earlier asset timeouts, readiness failures and unexplained later recovery remain separately recorded. No timeout, failure predicate or authority check was weakened. Focused public commands: ```sh pnpm exec vitest run server/src/__tests__/idle-sleep-preparation.test.ts server/src/__tests__/instance-settings-routes.test.ts --maxWorkers=1 --no-file-parallelism pnpm exec vitest run server/src/__tests__/heartbeat-reviewed-chat-binding.integration.test.ts --maxWorkers=1 --no-file-parallelism ``` Local database qualification uses fresh owned test databases on existing PostgreSQL, bounded pools and one worker. The local allocation adapter changes only database allocation for this test module; it runs normal migrations and client cleanup. It is not part of the published source. No new cluster is created. Local acceptance measured the `0a69efa6` integration with the same committed authorization bytes. The later normal master merge adds startup-timeout classification and recovery changes. Those paths are outside the controlled healthy-backend witness; retirement and authorization hunks are preserved. Exact-head remote CI at `7f6f02b611dcf5a9a2b05173a44e9631eaf78058` has completed with 56 successful contexts and four skips. The complete current Greptile summary is5/5 with no new actionable findings; all review threads are resolved. Ponytail’s current turn is explicitly paused without a current-head verdict; earlier actual findings are fixed and accounted for. No whole final-tree local repository build/test run, live-provider restore or deployed suspension is claimed. Actual master subsequently advanced to `d5c4f7dd02d408a261cedd9a1926b9900a4aedc4`. Its local normal-merge projection with this reviewed head preserves all20 owned paths and every unrelated master blob and mode. Current Runner TypeScript, App declarations and their full owning compiler programs pass. Focused Native cleanup and warm-continuation composition checks are being completed before merge; the new upstream suspension-retry behavior is not inferred from the older controlled-backend witness. The remote PR head remains unchanged. ## Risks - Privileged idle preparation can close idle warm Native sessions. It requires the exact current hold. - Failed closure retains its owner and can keep the tenant awake until normal retry succeeds. - A changed hold cannot authorize sleep. Started cleanup remains tracked until it settles. - The chat repair can perform one extra full authorization read. It keeps the existing deadline and does not grant authority from the fallback query. - Unsupported resources and active work retain their existing blockers. This PR does not certify every transport, plugin or deployed host. ## Model Used OpenAI `gpt-6.1-sol` produced the delegated changes and verification. OpenAI Codex, based on GPT-6, performed managing review and integration. Both used repository tools and code execution. The coordinator's exact backend snapshot and context-window limits were not exposed. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green at exact head `7f6f02b611dcf5a9a2b05173a44e9631eaf78058` - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups at the exact current head - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing> Co-authored-by: Codex <noreply@openai.com>	2026-11-06 20:49:30 -06:00
DottaandClaude Opus 5.5	a5fe5deebe	fix(environments): reclaim sandboxes left by a restart during acquire or teardown (#19696 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Remote agent runs execute in provider sandboxes (for example Daytona) that Paperclip creates and must clean up > - Before each provider create, Paperclip saves a durable creation intent (`sandbox_creation_operations`). Recovery uses it to find and destroy a sandbox whose lease was never committed > - Recovery waited 30 minutes after creation before it checked an intent, because a live server could still be waiting for the provider > - After a restart, the creating server is gone, but the intent still waited the full 30 minutes. A short-lived or redeployed instance never reclaimed those sandboxes > - The teardown that follows has the same problem: a teardown claim fences other sweeps for 15 minutes, and a restart during the claim kept the sandbox alive until the fence expired > - This pull request saves the server process identity with each intent and each teardown claim, and ends the wait when same-host proof shows that process exited > - The benefit is that a restart during sandbox acquire or teardown no longer leaves paid provider sandboxes behind ## Linked Issues or Issue Description What happened? A control-plane restart during Daytona sandbox acquire left the created sandbox without an `environment_leases` row. In one restart chaos wave, 6 creation intents stayed `pending` with `next_attempt_at` = creation + 30 minutes. The wave ended (and its environment was deleted) before that time, so 4 sandboxes stayed `stopped` in Daytona. Only Daytona's own auto-delete would remove them. Two of the runs did not exist in `heartbeat_runs` because they were non-run acquires; their intents still had the run id. Expected behavior After a restart, each sandbox that this instance and company created, with no committed lease and no live creator, is found by its exact creation name and labels and destroyed through the plugin, after the existing ownership checks. Sandboxes from another instance, another host, or a live server stay untouched. Nothing acts org-wide. Steps to reproduce 1. Start a Daytona-backed run. 2. Kill the server (SIGKILL) after the creation intent commits and before the lease commits. 3. Restart the server. On master, the intent stays `pending` for 30 minutes and the sandbox stays in Daytona. 4. The regression tests in `server/src/__tests__/sandbox-creation-operations.test.ts` reproduce this with the controlled provider fixture. Refs #18357 (added the durable creation intent this change completes). Paperclip version or commit master `07a548073a` (live proof at `8d525dba7d` + this change, content-matched to the runner image) ## What Changed - New `server/src/services/controller-process-identity.ts`: records the server boot id, PID, a PID-space hash and the process start time. The PID-space hash covers the hostname, the kernel boot id and the PID namespace on Linux (the hostname and boot time elsewhere), so containers that share a hostname but not a PID namespace never judge each other. It proves that a recorded process exited only in the same PID space: the PID is gone, or the PID is live with a definitely different start time. Another PID space, an unreadable identity, an unidentified record, or this server never counts as exited. - `sandbox-creation-operations.ts`: each creation intent stores the creating process and its settle deadline. Each recovery pass first moves intents whose creator exited to "due now". If the early lookup finds no sandbox before the settle deadline, the intent stays `pending` and retries. It does not become manual reconciliation, because the provider can still materialize an aborted request. - `heartbeat/recovery.ts`: each pending-cleanup teardown claim stores the claiming process, bound to its claim attempt id. Before the sweep reads due leases, it ends the 15-minute fence of a claim whose claimer exited. A newer claim from another path (workspace repair, export retry) that kept an older record has a different attempt id, so it is never released. - Both recovery passes page through candidates with a stable id cursor. Live, foreign or unidentified owners therefore cannot hide an abandoned record behind one page. - This covers every `createOwnedSandbox` caller, including non-task probe and qualification acquires. Their intents have no issue and a generated run id, which is why their run ids are missing from `heartbeat_runs`. - Recovery uses only the existing paths: exact `paperclip-create-<operation id>` name and label inspection in the plugin, `pending_cleanup` lease, and plugin teardown. No org-wide listing or deletion was added. ## Verification - Red, then green (controlled provider fixture, real embedded PostgreSQL): - `sandbox-creation-operations.test.ts` → "after the creating server exits mid-acquire". Before the fix, the exited-creator case left no cleanup lease. After the fix, the next sweep creates the `pending_cleanup` lease and the teardown destroys the sandbox. Counter-tests: a creation owned by this live server, another live process, another host, or a committed lease is not inspected or destroyed. An early "not found" keeps the intent pending, and a later sweep still finds a late allocation. - `heartbeat-pending-cleanup-sweep.test.ts` → "a teardown claim whose server stopped mid-attempt". Before the fix, a lost claim blocked the teardown. After the fix, it retries now. Counter-tests: claims from this live server, another host, or without an identity keep their fence. - Review regressions: "keeps the fence of a newer claim that kept an earlier attempt's server record" and "pages past more fenced claims than one sweep page" fail on the previous commit and pass now. "pages past more unprovable creations than one page" guards the creation-side cursor; the old unordered page happened to include the row, so it was not red. - After the review changes, `sandbox-creation-operations`, `heartbeat-pending-cleanup-sweep`, `live-recovery-work` and `environment-cleanup-work` pass (155 tests), and `tsc --noEmit` for `server` passes. - Related suites pass: sandbox-creation-operations (28), environment-cleanup-work, task-creation-recovery, native-startup-cleanup-recovery, sandbox-orphan-cleanup-spool, plugin-warm-tasks, idle-completed-history (187 total), and the pending-cleanup sweep suite. `tsc --noEmit` for `server` passes. - Live Daytona restart chaos wave `w-20261107T001609-s11` (codex, A1, K2+K4, concurrency 2, RC-06 masked, image `sha256:2767e91c…`, first fix only): 4/4 tasks pass, all restart-crossing, R6 pass for every task, `orphanSandboxes: []`. The server log and database show two creation intents killed by K4 being reconciled right after restart and destroyed through Paperclip. A third reclaimed sandbox lost its teardown claim to the next K4 kill in the same second, so its 15-minute fence blocked the environment delete. The second commit fixes that. - Live Daytona wave `w-20261107T012048-s11`, same plan, both commits: 4/4 pass, all restart-crossing, R6 pass for every task, `orphanSandboxes: []`, no held leases. The environment delete through Paperclip succeeded with `remainingSandboxes: []`. One restart-time sweep destroyed one sandbox through the plugin. ## Risks - Same-host proof relies on PID liveness and process start time, as native controller takeover already does. A false "exited" result is limited to recovery running early. The plugin still checks exact name and labels before any delete. An early not-found result stays pending. - Multi-host deployments sharing one database keep the old deadlines, because a different PID space is never judged. A redeploy into a new host or PID namespace also keeps the old deadlines (fail closed). - The live waves ran before the review commits. The review commits only narrow when a record is judged, and they add paging. - Intents and claims written before this change have no process identity, so they keep the old deadlines. - No schema change. The new fields live in existing JSON metadata. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file editing, test execution). Extended reasoning enabled. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 20:39:34 -06:00
DottaandPaperclip	d5c4f7dd02	fix: preserve shared Git sources during online workspace migration (#19688 ) ## Thinking Path > - Paperclip preserves the work produced by its agents. > - Remote workspace migration must preserve legacy Git history and local edits. > - Linked checkouts can share administration and objects with other tasks. > - A task lock alone cannot exclude writers of a shared or borrowed store. > - The online source owner must derive and retain the full physical ownership closure. > - This change reuses task, project, lifecycle and finalizer locks for that closure. > - Idle legacy Git work can then migrate while unrelated remote turns continue. ## Linked Issues or Issue Description Refs: #19642. What happened? Online migration keeps legacy Git sources blocked because several task checkouts can share a project base or a private Git cache. A read-only source observation found 160 pending Git records, 120 distinct paths and nine common stores. Its largest project cohort contained 126 records and 110 task owners. An ordinary-directory fence cannot preserve these sources by itself. Expected behavior Derive the physical source owners from stable Git metadata. Hold their existing source and lifecycle keys through publication. Retain all history and recovery obligations. Reject foreign aliases, unknown stores and changed sources. Let unrelated object-backed work continue. Steps to reproduce 1. Create a managed project checkout and linked task worktrees. 2. Keep a settled prepared recovery source on a sibling task. 3. Keep unrelated remote work running. 4. Request an online import of an idle leaf checkout. 5. The existing owner keeps the shared source blocked. Paperclip version or commit This branch builds on merged #19642 (`7ac99718a4`) and fresh master `fb3cf1a072`. It adds the Git closure. It does not add a writers-stopped shortcut. Deployment mode Hosted native Runner with remote workspace storage. Local qualification uses owned synthetic Git repositories, disposable PostgreSQL databases and the actual archive and storage owners. ## What Changed - Read bounded stable Git pointers, common-directory metadata and local alternates. Derive the complete physical source ownership cohort. Preserve all object stores. - Retain common and borrowed store owners in preparation, lifecycle, runtime, expiration and actual reopen paths. Keep company boundaries and the existing 128-key bound. - Hold exact sibling native recovery proofs under their existing physical finalizer keys. Recheck descriptors, allocations, recovery actions and source identities before capture and publication. Preserve original states and warnings. - Require leaf-first nested imports. Reuse the existing immutable child download, restore and source comparison before parent exclusion. - Bind the self-contained archive owner to the held Git namespaces. Keep its three walks, history normalization, fsck, bytes and digest format. Reject bare, detached, network, unknown or changed Git storage. - Check both foreign path fields. Keep foreign metadata private. Document the proof, bounds, source observation and remaining eligibility limits. ## Verification - Final revision `2934e41c923b93e16508b978c849b747c86be645`: all checks are terminal, with 56 successful and four skipped, zero failures and zero pending. The [required CI aggregate](https://github.com/paperclipai/paperclip/actions/runs/42326464408/job/128266279210) passed. Build, types, all server/Runner/serialized suites, end-to-end aggregate and Snyk passed. [Fresh exact-head Greptile review](https://github.com/paperclipai/paperclip/pull/19688#issuecomment-6699760104) is 5/5 with no open threads. GitHub reports MERGEABLE and CLEAN. - Review then found two correctness gaps: an ordinary managed directory inside an unrelated host repository was blocked, and a Git parent could miss a completed nested operation after its child snapshot became missing or deleted. Three actual-owner regressions reproduced both gaps in 9.70 seconds. The corrected owner passes 12 focused cases in 17.08 seconds. It imports and restores only the ordinary root bytes, preserves the outer repository, rejects both missing/deleted child receipts before a parent journal, and retains valid/changed child coverage and private-cache contention. Its wrapper exited 0 and cleanup completed. The first bootstrap attempt failed before tests ran because the host exhausted its 32 shared-memory segment limit; no shared services were stopped and no pass is claimed for that attempt. The final qualified source/test type graph, module boundaries and diff checks pass. - Receipt scans keep canonical path resolution. A lexical SQL prefix can omit children reached through symlinked ancestors. The observed full metadata closure completed in 1.17 seconds; no measured bottleneck justifies weakening that alias proof. Writer admission still rejects uncertain Git ownership before mutation. Expiration returns its typed skip, and reopen returns its structured failure. Runtime and lifecycle do not bypass an unknown source. No broad error-contract change was added. - Full CI at `720104ea975e` exposed five existing boundary cases across three server shards. Four exact cases reproduced locally as failures in 25.98 seconds. Reopen now preserves its structured failure when source inspection rejects the path, without reconstructing it. Runtime recognizes only the exact regular ignore-only `.git/info/exclude` fragment; migration still rejects that fragment. The original arbitrary-path and file-URL import/delete-host/resume fixtures keep both supported input forms and use the genuine offline setup path after their sole writer finishes. The production mode gate is unchanged. A new real-owner control proves the offline flag still rejects a running legacy database writer. - The final focused run passed all ten cases across four suites in 22.75 seconds, including all five CI failures, invalid-pointer admission, fragment rejection, offline writer exclusion and actual nested reopen lock ordering. Its wrapper exited 0 and cleanup completed. The known invalid-pointer case remains fail-closed; the separate unexpected-I/O diagnostic retains safe error metadata without command text or credentials. Existing Git stderr redaction coverage remains intact. The final metadata reader suite passed 20 cases in 2.68 seconds. The qualified server/test source type graph, module boundaries and diff checks pass. No production assertions or timeout defaults were weakened. Fresh CI and review are required for the repair revision. - A later review found that the ordinary early return still used only one foreign path field. The actual ordinary providerRef-only regression failed as incorrectly eligible. One source-path owner now enumerates both local locators for company/foreign ordinary and Git scans, layout checks, early admission, lifecycle/reopen/runtime and expiration/move/preparation retention. Capture and display keep their original preferred path. The actual owner canary passed 49 cases across two suites in 109.15 seconds. It covers cwd-only/ref-only/different/symlink locators, same-company mismatches, foreign owners, lifecycle ordering, Git/recovery imports and expiration/reopen. Three selected-locator fallback cases also failed before their guard; the final eight-case control passed in 31.05 seconds, including the configured alias and valid import/private-child controls. All wrappers exited 0 and completed cleanup. The metadata/locator suite passed 19 cases. The final source/test type graph, module boundaries and diff checks pass. - Review found two P1 gaps. Both configured-directory symlink cases failed before canonicalization. The actual parent import lost an already-imported child's unique private cache and failed its archive fence instead of retaining that owner. The corrected real-owner canary passed four cases in 20.78 seconds: configured symlink runtime provisioning, private-cache contention before the parent journal, successful parent publication after release, and unchanged/changed child coverage. The metadata suite passed 15 cases in 4.14 seconds. Both wrappers exited 0 and cleanup completed. - A fresh source read-only observation found 160 pending Git records and nine common stores. An additional exact receipt query found 1,243 completed migration receipts and 557 nested source paths under pending Git roots. All 126 records in the largest project were present in the captured Git metadata set. For all 110 present candidate workspace records, the derived closure fits the unchanged 128-key cap. The worst case needs 110 issue keys and one project key, with 17 keys of headroom. Its 173 imported children introduce no physical issue owner outside those 110 logical task owners. This checks the observed key bound only. It does not claim migration eligibility or a coherent snapshot. - The initial stacked CI did not run four Runner test lanes: trusted master passed new --package-lane/--native-server options to the old base lane script. The branch is now rebased onto actual merged master and retargeted before synchronization. It includes the canonical current lane parser without an unrelated Runner patch. Fresh new-head CI is required. - The actual importer/source owner run passed 36 cases across three suites in 187.75 seconds. It covers sibling prepared recovery, unrelated running remote work, common-cache contention, changed coordinator/allocation/path state during upload, exact nested coverage, dirty unpushed work, and existing expiration/reopen owners. Wrapper exit 0 and cleanup completed. - A separate previously hanging lifecycle-worker restore passed in 17.33 seconds with the unchanged 15-second test-body limit. An earlier run timed out in three expiration/reopen cases and blocked cleanup; its wrapper later exited 1. These failures remain recorded. They are not counted as passes. - Two boundary controls first failed: a foreign providerRef-only alias was incorrectly eligible, and actual reopen did not admit its distinct cache owner at the initial source boundary. The narrow guards now inspect both foreign locators and derive physical Git keys before the raw lifecycle transaction. The foreign alias and existing real nested reopen contention passed. The private-cache control first retained contradictory recorded source provenance and stayed fail-closed after release. The corrected present legacy-worktree fixture passed in 12.61 seconds. These failed attempts remain recorded; they are not passes. - The metadata reader and serialized/local archive controls use actual Git repositories. They compare archive bytes and digests, remove every original source, restore staged and unstaged work, and run fsck. The metadata reader passed 13 cases in 2.85 seconds. The local/serialized archive controls passed nine cases in 9.27 seconds. All processes exited 0. An additional actual importer control changed a relative commondir pointer to the absolute name of the same store during upload. Publication was rejected, the uploaded journal remained available, and original files stayed intact (one case, 5.86 seconds; wrapper exit 0 and cleanup complete). - The final rebased full local server/test type graph passes with db/shared/Runner source aliases because reused dependency declarations predate new master exports. The first rebased attempt reported those stale declaration fields/tables, and is not counted as a pass. Module boundaries and diff checks pass. Standard package builds, types and broad regression checks remain required in CI. - No hosted import, admission hold, cancellation, provider mutation, source filesystem write or shared-cache Git command was performed for this change. The source observation is read-only and is not a locked eligibility proof. ## Risks - A missing writer owner could make physical source exclusion incomplete. Existing mutation owners retain the same company/task/project keys, and uncertain metadata fails closed. - A project can contain a large cohort. Limits remain finite. A bound failure keeps the source blocked. - Bare, detached, unowned or network Git stores remain blocked online. Missing archived sources retain their existing authenticated reconstruction path. - Exact recovery proofs preserve authority and obligations. They do not infer provider termination or mark a prepared descriptor finalized. - This change does not claim all observed Git records are immediately eligible. Existing preparation, provider, recovery, runtime and source-change blockers remain. ## Model Used OpenAI GPT-6 through Codex, with reasoning, tool use and code execution. The host does not expose a more specific backend model ID or context-window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 20:08:19 -06:00
DottaandClaude Opus 5.5	d26372dc8a	test(restart-chaos): add a result trigger and recovered-run comment-wake guards (#19689 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - The restart chaos harness kills the control plane at chosen moments and grades whether each task still finishes correctly > - One wave lost a user's follow-up comment after a kill that landed after the run accepted its result but before the run settled > - No existing trigger aims a fault at that moment, so the case was hard to reproduce on purpose > - The recovery path that settles such a run had no regression test that covers the saved comment > - This pull request adds a `result` trigger to the harness and a regression test for the recovery path > - The benefit is that the harness can aim faults at the result-accepted window, and a change that drops, repeats, or re-wakes a saved comment fails a test ## Linked Issues or Issue Description What happened? In a restart-chaos storm wave, a user posted a comment while a native run was active. A hard kill (K1) landed after the run accepted its result and before the run settled. The run settled `succeeded` and the task stayed `done`, but no follow-up run read the comment. Steps to reproduce The original failure did not reproduce. Three targeted waves ran 12 tasks with K1 at comment time or in the new result window. Every follow-up comment was consumed exactly once. The new regression test also passes on current `master`. The test and the trigger guard the path and do not claim a fix. Expected behavior A comment posted during a run is consumed exactly once after the run ends, including when restart recovery settles the run. ## What Changed - `tests/restart-chaos`: add `--triggers result` (C1 only). K1 fires 0–3 s after the run that received the follow-up proposes or accepts its result (`run.result.proposed` or `run.result.accepted`). The fault fires only while that run is still `running`: the observer checks the latest sample and rereads the runs right before dispatch. A run that settled during the delay, or a follow-up posted at rest, records a `result` miss. Add poll-driven unit tests for these cases and a README row. - `server/src/__tests__/native-run-projection-work.test.ts`: add embedded Postgres tests in which background recovery, not the executor, settles the run (projection, environment settlement, and deferred-receipt workers), with a deferred user comment and an active local lease left by the killed executor. The tests check: - The comment is promoted into exactly one follow-up run, and a second restart adds none. - A comment the run already received as a same-turn steer is not woken again. The other saved comment still is. - A queue whose only comment was steered stays cancelled, and no follow-up starts. - A Retry click does not consume the comment a second time. A click that observed the source is stale. A click that observed the follow-up joins it. ## Verification - `cd tests/restart-chaos && npx vitest run --config vitest.config.ts` (96 tests pass) - `cd server && npx vitest run src/__tests__/native-run-projection-work.test.ts` (54 tests pass) - Live, local, ACPX Claude, Codex and OpenCode: `node cli/node_modules/tsx/dist/cli.mjs tests/restart-chaos/main.ts --env local --workloads C1 --triggers comment --faults K1 --adapters codex,claude,opencode --tasks 4 --concurrency 2 --rc06-masked` passed 4 of 4. Two waves with `--triggers result` consumed each follow-up exactly once. One task in those waves found a separate defect: a kill 1.4 s after the result commit left the run `running` behind the dead controller's 20-minute finalization lease. A separate pull request fixes that defect. ## Risks - Low risk. These are test and harness changes only. No production code changes. - The `result` trigger depends on the observer's poll interval (about 3 s), so the exact kill point within the window varies. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, run as an agent in Claude Code with tool use (shell, file editing, local test execution). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 19:53:58 -06:00
DottaandPaperclip	5e4d24b6c8	fix(recovery): preserve admitted workspace on native restart (#19693 ) ## Thinking Path > - Paperclip manages AI agents and their work. > - A Retry can admit a new workspace before native provider work starts. > - The native execution input records that admitted workspace. > - A server restart must recover the same run under that recorded contract. > - Retry readiness previously evaluated fresh admission again after the run had started. > - This could replace the workspace and make the retained lease fail its correct identity checks. > - This change keeps the admitted workspace and leaves reattachment authority with the existing lease and controller checks. ## Linked Issues or Issue Description What happened? A native run started by Retry could do useful work, then fail after a server restart. Preparation evaluated the original fresh-workspace receipt again. It reserved a replacement workspace under the Retry intent ID. The existing runner still owned the original workspace, so setup failed with `native_remote_recovery_lease_mismatch`. Expected behavior Restart recovery must retain the workspace recorded in the same run's native execution input. A later configuration or admission decision must not replace it. Steps to reproduce 1. Admit a remote workspace for a Retry run. 2. Persist its native execution input and keep its original controller and environment lease records. 3. Recover the same run while its canonical Retry receipt selects fresh admission. 4. Preparation must reach the original lease with the original workspace. Related: #19340 preserves verified Retry workspaces. This change handles an already admitted native run instead of selecting a workspace for a new turn. Public duplicate searches for admitted/fresh workspace restart found no matching repair. ## What Changed - Stop fresh admission when a parsed native input names a real workspace in the run's company. - Validate the company, run, agent, and task binding before using that input. - Keep synthetic run-ID bootstraps on the existing admission path. Hold missing real bindings. - Test both recovery-action and ordinary user-Retry receipts against real workspace, controller, and lease rows. Check malformed, foreign, missing and opaque bindings, synthetic bootstrap behavior, and no-files API controllers. - Align the resume-owner fixture with the real transient run-ID binding. Give each invocation a unique issue label so a CI test retry cannot mask the original failure. ## Verification - RED on unchanged production: the new reservation regression failed with `native_remote_recovery_lease_mismatch` instead of reaching the admitted lease. - Review RED: an opaque task-workspace ID reached the UUID query. The new guard returns the controlled missing-admission error before the query. - Review RED: a valid no-files Cursor controller was treated as a missing task workspace. The existing task-workspace projection now keeps that controller on its prior path. - Clean final production-source regression: both restart cases passed in a fresh embedded PostgreSQL cluster. These cover both receipt paths and all binding, bootstrap and API-controller controls. - The initial focused checks passed 11 cases. Seven existing ordinary Retry and environment selection cases also passed. A later combined final-base run had 17 assertions pass but failed workspace teardown at 90 seconds and task-Retry setup at 30 seconds. That attempt is retained as a failure; the two new restart cases then passed alone with exit 0. Earlier PostgreSQL support-probe failures are also retained as setup failures. - The worktree's own Runner, shared and SDK dependencies were built. All seven server typecheck layers passed for the final production source. The final additional commit changes only the existing resume-owner fixture. - Required CI found an invalid fixture binding in [server shard 11](https://github.com/paperclipai/paperclip/actions/runs/42322379917/job/128250151095). The first attempt failed before lease acquisition because the fixture named a random workspace with no row. The automatic test retry then failed on its fixed issue label. The fixture now uses the transient run-ID contract and a unique label. Its exact original acquisition and owner-release assertions pass locally with exit 0. - Whitespace and added-code-comment checks passed. - Final commit `f0e41e8c6b949f43863ec60fc7b96b97025eafc6` passed required [CI](https://github.com/paperclipai/paperclip/actions/runs/42323826555): 56 successful checks, four intended skips, one neutral result, and no pending or failed checks. Greptile and Ponytail both gave this exact head 5/5 with no actionable findings. No review threads remain open. Hosted deployment and original-task recovery remain separate from local proof. ## Risks Malformed or missing native admission records now hold rather than selecting fresh work. Valid initial admissions and synthetic run-ID bootstraps retain their prior path. The change does not grant reattachment authority or settle provider effects. Existing controller, workspace writer, lease, Stop, and trust checks still apply. No API, schema, or configuration contract changes. ## Model Used OpenAI GPT-6 family through Codex. The runtime does not expose the exact serving model ID or context window. No model override was used. The assistant used reasoning, code editing, shell tools, and local tests. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 19:37:48 -06:00
DottaandPaperclip	b94963f91e	fix: retry temporary repository catalog failures before dispatch (#19690 ) ## Thinking Path > - Paperclip manages AI agents and their work. > - A task must check repository access before it starts agent work. > - Repository discovery can fail while the user's existing grant remains valid. > - Discovery previously replaced a known temporary provider refusal with an empty catalog. > - The task then reported lost access and did not schedule recovery. > - This change keeps temporary discovery failures distinct and uses the existing bounded retry owner. > - Each retry must pass a fresh access check before it can use the saved repository. ## Linked Issues or Issue Description What happened? A task with a saved catalog repository stops with an access denial when GitHub returns an exhausted-quota response or HTTP 503. The task does not start provider work, but the discovery layer discards the temporary failure and returns an empty catalog. Expected behavior Known temporary failures before provider dispatch use the existing bounded infrastructure retry. The successor checks the same responsible user's current grants and reads the saved files only after fresh discovery succeeds. A completed denial remains terminal. Steps to reproduce 1. Bind a task to an authorized catalog repository with saved local changes. 2. Make its authorized GitHub discovery return HTTP 403 with zero remaining quota, or HTTP 503. 3. Let discovery recover before the existing retry deadline. 4. Observe one successful successor using the original repository bytes. Paperclip version or commit Reproduced on master at `b8d94b3ad1`. A search found no existing PR for this temporary catalog failure. ## What Changed - Keep a safe typed failure for known HTTP availability and quota responses. Honor the later provider deadline and bound it to one day. - Propagate that failure only when fresh discovery cannot find a required repository. Preserve successful partial catalogs and the existing connection setup error contract. - Classify the failure only before provider dispatch. Reuse existing bounded retry, interaction ownership, and bootstrap settlement rules. - Keep current grant checks, saved repository selection, normal Stop, and provider-started failure policy. - Keep provider deadlines when accepted interactions use their own retry owner. Recognize secondary HTTP 403 rate limits with valid retry advice while keeping ordinary permission denials terminal. This matches [GitHub rate-limit guidance](https://docs.github.com/en/rest/using-the-rest-api/rate-limits-for-the-rest-api). - Keep direct repository request responses outside the automatic bootstrap retry contract. Test successful partial discovery and missing-required-repository recovery with two independent eligible connections. - Document task repository discovery recovery. ## Verification - Behavioral RED: real heartbeat and PostgreSQL fixtures reached actual HTTP 403 and503 discovery responses, started no adapter, and found zero successors instead of one. Both cases failed before the production change. - Behavioral GREEN: both cases waited for the real retry deadline, used the normal queue worker, ran the controlled adapter once after fresh discovery, and read the original saved bytes. Two cases passed in 74.73 seconds. No external model or provider request was made. - Deadline RED/GREEN: a combined Retry-After and quota-reset response initially selected the earlier deadline. The regression now selects the later deadline. - Initial repair at `d8d4714a18` (same production source as `018e8c52d345d79972d7907468080dc9908c299f`): seven affected files passed, 69 cases in 185.16 seconds. That test-only cleanup then passed all five real-database acceptance cases in 134.76 seconds. - Commands: `pnpm exec vitest run` with `heartbeat-repository-catalog-recovery.test.ts`, `github-catalog-timing.test.ts`, `github-grant-metadata.test.ts`, `project-repositories-persistence.test.ts`, `task-workspace-repositories-persistence.test.ts`, `github-repository-catalog-failure.test.ts`, and `services/heartbeat/retries.test.ts` from the server source tree. The acceptance file also ran fresh at `018e8c52d3`. A fresh five-case run after the P1 repairs passed at `f490ca6b78` in 122.06 seconds. The final eight-case file passed fresh at `e5b2efab11919467850f55fd6345317132b9b672` in 124.62 seconds, using real owned embedded PostgreSQL. It covers automatic recovery with saved bytes, genuine denial, provider-started classification, normal Stop, the direct HTTP route, and both partial-catalog outcomes. - Review repairs at `f490ca6b787cf77df96d3d6edfb2ab9115eb19a6`: both reported P1s reproduced RED. The real database interaction scheduler chose a retry after 24 seconds before the provider deadline of one hour. A secondary HTTP 403 with valid Retry-After lost its retry classification. The narrow fixes then passed 44 cases across scheduler, header, and metadata-contract files in 37.66 seconds. - A real direct request through the task workspace router and normal error middleware reproduced a 500 response instead of the existing 403. Required-ID retry mode now applies only to admission and its rechecks. The route and both two-connection partial-catalog cases then passed in 29.09 seconds. Both partial cases use actual owned grants and credential bindings. - All seven server type layers passed again at final head `e5b2efab11919467850f55fd6345317132b9b672`. The affected test type layer passed again after the test-only cleanup. Environment: macOS arm64, Node 26.4.0, pinned pnpm 9.15.4, Vitest 5.0.3, owned disposable PostgreSQL 18 on loopback. The current checkout's shared packages, plugin SDK, and Runner were built first. - One local acceptance run under concurrent owned test/typecheck load hit the unchanged 15-second Stop-case timeout. Its four other cases passed. The unchanged complete file then passed serially. The timeout receipt remains recorded; no assertion or timeout was weakened. - A provider-started failure retains the existing generic adapter retry policy. It does not receive the catalog no-provider certificate or a provider retry deadline. Fixture cleanup cancels its scheduled retry through the normal owner. - Final head `e5b2efab11919467850f55fd6345317132b9b672` has 59 successful checks, 3 allowed skips, no pending or failed check, and Greptile 5/5 with no open findings. [Completed CI workflow](https://github.com/paperclipai/paperclip/actions/runs/42322202977). - Additional local diagnostic for the late recovery-owner advisory: on unchanged `e5b2efab`, the dedicated scheduler returned `retry_exhausted` with attempt 3/max 2. The normal stranded sweeper then escalated once, created no extra run, and started no adapter, even after catalog access recovered. One case passed in 17.32 seconds. The existing failed-predecessor path delegates to that bounded scheduler and returns; it does not enter a new generic retry budget. No production registry change was made. This diagnostic was not a CI case. Its local harness is preserved separately at `74e3854eea3f7c4121fdcd9c8a8ad522cd8c7477` on `codex/catalog-exhaustion-acceptance`; its production source is identical to `e5b2efab`. - Module boundary and whitespace checks passed. ## Risks - A temporary failure from another eligible connection can delay a truly missing repository for the existing limit of two attempts. It cannot grant access. - Unknown network errors and genuine access denials retain their existing behavior. This change handles known HTTP failures only. - Historical provider quota causality is unproven. This repair has a separate local producer regression. It does not rewrite past failures or queued user intents. - Acceptance uses real local database and heartbeat owners with controlled HTTP and adapter fixtures. It does not claim a live native-provider recovery. ## Model Used OpenAI GPT-6.1 Codex (`gpt-6.1-sol`), with high reasoning, repository tools, code execution, and controlled local PostgreSQL tests. The host did not expose a context-window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 19:29:40 -06:00
DottaandClaude Opus 5.5	06bbcea12f	fix(budgets): price lost receipts and stop restart-derived accounting from hard-stopping companies (#19673 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Budget admission and run accounting decide whether any task in a company may start > - #19497 made pending accounting an admission hold and settles a stopped run's lost receipt as an unpriced charge > - Restart-chaos waves still stopped whole companies. A lost Codex receipt with complete token counts settled as unpriced because the rate card only prices complete receipts. Under the default blocking policy that hard-stopped the company at 0.38 of 3000 cents. In another storm, a resumed run's continuation was refused by its own task's recovering-accounting hold. Separately, unpriced charges with zero tokens or a known partial cost hard-stopped a company at 120 of 2000 cents > - This pull request prices a lost receipt's observed usage from the rate card, lets a task's continuation pass its own recovering accounting, and stops restart-derived or evidence-free unpriced charges from tripping the blocking policy > - The benefit is that a restart no longer pauses a company by itself, while genuinely unpriced usage still blocks ## Linked Issues or Issue Description Follow-up to #19497. Refs #19606. What happened? - Wave A (K6 crash loop): native Codex run `5037b8d1` was killed mid-turn. Its receipt was lost, but `usage_json` held gpt-6.1-sol with 28690 input and 220 output tokens. The charge was written `cost_status=unpriced`, `pricing_provenance.source=unknown`, because `priceCodexReceipt` returns early when `!receipt.complete`. 0.2 s later the default `unpricedUsagePolicy=block` opened a hard incident at 0.38 of 3000 cents. The company was paused, and 6/6 tasks were blocked with `canRetry=false`. - Wave B (K1 storm): Claude run `2a7a8abf` resumed after the restart. `reserveRunBudget` refused its interaction continuation with "New work must wait while a native run recovers its unfinished accounting", so the task could not settle its own recovery. A 0-token unpriced charge and an unpriced charge with a known 0.81-cent cost then hard-stopped the company at 120 of 2000 cents. What did you expect? A restart-lost receipt with known numbers is priced as an estimate, and no restart-derived accounting state pauses a company by itself. A task's own continuation is never held by its own accounting. Real usage above the limit, and genuinely unpriced usage, still stop work. Steps to reproduce 1. Give a company a hard-stop budget with the default unpriced-usage policy. 2. Kill a native Codex turn after it has reported tokens, so the final receipt is lost and native recovery ends in terminal failure with a verified stop. 3. Observe an unpriced charge for known tokens, a hard incident, and a paused company. ## What Changed - `run-cost-accounting.ts`: when a lost receipt settles and its usage has known tokens and no known cost, price it with `priceCodexReceipt`/`priceAnthropicReceipt` as a complete per-run checkpoint. The charge is `estimated`, its provenance keeps the rate-card source, and its evidence says "final receipt lost; observed usage is a lower bound". The estimate is also written to the run's usage so a later receipt reconciles against it. Zero observed tokens, and per-model usage, stay unpriced (unknown is never shown as zero, and aggregate totals never take one model's rate). - `budget-exposure.ts`: - The pending/recovering accounting hold excludes the run being admitted and earlier runs of the same task (same `native_issue_id`). A task's continuation and finalization can therefore settle its own accounting. Other tasks stay held, and those runs still count as spend. - The blocking unpriced-usage policy no longer counts these unpriced charges: the settled final charge of a lost receipt (restart-derived, any usage; later growth still counts), a charge with a known partial cost, and metered API or subscription billing with no observed tokens. Unknown billing with no tokens (for example Cursor Cloud) and complete-receipt usage that has no price still block. - `budgets.ts` / `budget-reservations.ts`: admission accepts the admitted `runId` and passes it to the exposure read. `reserveRunBudget`, the final dispatch recheck and the read-only admitted recheck all supply it, so a continuation that reserves can also dispatch. - Daytona wave s12 (run f1d98917, a blocked remote recovery with a 0-cent held reservation and every lease destroyed) is closed by #19606. A Retry for that same task passes the hold through the same-task exclusion here. ## Verification - `closed-native-receipt-accounting.test.ts` (embedded Postgres): - Lost Codex (gpt-6.1-sol) and Claude (claude-sonnet-5) receipts with observed tokens settle as `estimated` rate-card charges with the lower-bound evidence. The company stays active and admission is clear under the blocking policy. - A lost receipt for a model with no rate card stays unpriced but opens no incident and no pause. - A task's continuation reserves budget past its own recovering accounting, while another task's run is still held. - Unpriced charges with no tokens on metered billing, or with a known partial cost, do not hard-stop. - Complete-receipt usage with no price still blocks. - The six new cases fail on master and pass with this change. - Existing budget and accounting suites pass locally: budget-enforcement-scoping, budgets-service, heartbeat-cost-accounting, run-cost-accounting, native-stop-accounting, cost-data-quality, budget-live-attempts, budget-live-usage. Server typecheck passes. ## Risks - A lost receipt's estimate is a lower bound. The provenance says so, and a later complete receipt still reconciles. - Restart-derived unpriced charges no longer trip the blocking policy. Usage lost with a killed provider is not counted against that policy, which is consistent with budget enforcement being best effort. Unknown-billing and complete-receipt unpriced usage still block. - The same-task exemption relies on `native_issue_id`. Legacy runs without it keep the old behavior. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file editing, embedded Postgres tests). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 18:55:26 -06:00
DottaandClaude Opus 5.5	29e05f7e06	fix(heartbeat): sweep orphaned run scratch directories (#19671 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Each heartbeat run gets a scratch directory in the system temp folder > - Each process attempt makes a new directory with a marker file > - Cleanup removes only the directory of the current attempt > - A restart kills the old attempt, so its directory stays forever > - One restart storm left 83 directories across 78 runs > - This pull request sweeps those directories when the run ends and at startup recovery > - The benefit is that temp space no longer fills with leftover directories ## Linked Issues or Issue Description What happened? Run scratch directories made before a restart are never removed. A restart storm left 83 `paperclip-run-` directories across 78 runs. Each holds only `.paperclip-run-scratch.json`. Steps to reproduce* 1. Start a run that makes a scratch directory. 2. Kill the server process and start it again. 3. Let the run finish. 4. List `$TMPDIR/paperclip-run-`. The old directory is still there. Expected behavior* The old directory is removed after the run ends, or at startup. ## What Changed - Add `sweepOrphanedRunScratch` in `server/src/services/run-scratch.ts`. - The sweep removes a directory only if it has a valid regular-file marker and its run is terminal in this instance database. It keeps a directory whose run it cannot see, because another instance can share the temp root. - The sweep keeps active runs, unmarked directories, symlinks, and directories outside `paperclip-run-*`. - The sweep checks each directory again before removal and bounds the scan (500, with a rotating start so kept entries do not starve later ones) and removals (100). A per-run sweep filters by directory name first. Marker run ids that are not UUIDs never reach the database query. - Call the sweep after the current attempt's cleanup in `heartbeat.ts`, for that run only. - Call the sweep in `reapOrphanedRuns` (startup recovery) with run status from the database. ## Verification - `pnpm --filter @paperclipai/server exec vitest run src/services/run-scratch.test.ts` - New tests cover: leftover directories of a terminal run are removed, an active run is kept, an unmarked directory is kept, a symlink is not followed, and removals are bounded. ## Risks - Low risk. The sweep acts only on marked directories of terminal runs. - A directory left by a previous attempt could hold a live grandchild process. The sweep does not check old process groups. The run is already terminal. ## Model Used - Claude, claude-opus-4-5 class model via Claude Code, with tool use and code execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 18:51:31 -06:00
Nicky LeachandClaude Opus 5.5	0b73c805e8	feat(environments): fill an unset instance default with a managed plugin environment (#19551 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - A managed instance gets its instance-level environments from the `PAPERCLIP_MANAGED_CONFIG` document, which can declare plugin environments next to the managed sandbox > - A platform that runs managed instances wants a new instance to start with one of those plugin environments as its default, without changing the default of any instance that already has one > - #19632 made boot fill the instance default only when it is unset, local or deleted, in one locked transaction, and removed the revert, so every existing managed instance keeps a set default > - This pull request adds an `instanceDefault` flag to a plugin entry. That entry fills an unset default through the same locked fill, before the managed sandbox does > - The fill follows the same readiness rule as an admin's default write, so it never moves ready agents onto a plugin that has not prepared them > - The platform also needs a way to know whether a release understands plugin entries, so the release now ships a small capabilities file > - The benefit: new instances get the plugin environment as their default, and existing instances and admin choices are never changed ## Linked Issues or Issue Description Refs #19632 (the locked fill-if-unset rule this builds on), #19508 (plugin entries in the managed config) Subsystem affected Server environments: the managed config parser and the managed environment bootstrap. Problem or motivation The managed config can declare plugin environments, but it can't make one the instance default on a new instance. An external platform also has no way to tell whether a given release supports plugin entries before it sends one. Proposed solution - Add `instanceDefault: true` to at most one plugin entry. Right after that entry's boot ensure succeeds, it fills the instance default through `fillUnsetInstanceDefault`. The managed sandbox fill runs after it. - On a new instance the plugin environment wins. On an existing instance the default is already set, so neither fill changes it. - The fill declines while an active agent that follows the default hasn't been prepared by that plugin. This is the rule the instance-settings route enforces. When the plugin fill declines, the managed sandbox fills as usual. - Ship `server/src/services/managed-config-capabilities.json`, which lists `pluginEnvironments`. Alternatives considered - A recorded one-time decision with a retry worker. It needed pending states and extra checks for many writer and trigger combinations. Filling only an unset default, under the existing lock, needs none of that. - Rejecting the fill whenever agents exist. That is stricter than the admin route and blocks harmless cases, such as agents that are paused or already prepared. Roadmap alignment This extends existing managed-environment behaviour. It adds no UI. ## What Changed - `server/src/services/managed-config.ts`: - `ManagedPluginEnvironmentSpec.instanceDefault` (boolean, default `false`). - The parser refuses `instanceDefault` on more than one plugin entry, on the sandbox entry (even `false`), or as a non-boolean. - `MANAGED_CONFIG_CAPABILITIES` and `MANAGED_CONFIG_CAPABILITIES_SCHEMA_VERSION` constants. A plugin entry is refused unless `pluginEnvironments` is listed. - `server/src/services/managed-config-capabilities.json`: `{ "schemaVersion": 1, "capabilities": ["pluginEnvironments"] }`. A test pins it to the constants and to this literal contract. - `server/src/services/managed-environments.ts`: - New `fillPluginInstanceDefault`. It runs right after a successful boot ensure of the flagged entry and logs a fill or a decline. Recovery after boot never fills. - The managed sandbox fill now runs once, after the plugin entries. - `server/src/services/managed-instance-default.ts`: `fillUnsetInstanceDefault` runs `agentsNeedSetupForEnvironmentChange` inside its locked transaction and returns `declined: "agents_need_setup"`. A sandbox target never needs agent setup, so the managed sandbox fill behaves as before. - `server/src/services/plugin-environment-driver.ts`: the readiness query is factored into `agentsNeedSetupForEnvironmentChange`. `assertAgentsReadyForEnvironmentChange` wraps it and still throws the same conflict. - `doc/DEVELOPING.md`: documents `instanceDefault`, the fill order, the readiness decline, and the capabilities file. ## Verification - `cd server && npx vitest run src/__tests__/managed-plugin-environments.test.ts` (embedded Postgres), under `instanceDefault (fill only an unset default)`: - A new instance gets the plugin row as its default. - An existing sandbox default is kept, and so is an admin's choice. - If the plugin was down at the first boot and the sandbox filled, a later recovery and boot keep the sandbox default. - An admin-cleared or local default is declined while a ready or verifying agent follows the default without the plugin's setup. The managed sandbox then fills it. - It fills when the agents that follow the default are paused or already prepared, or are pinned to another environment. - Without the flag, nothing happens. - A new agent that follows the default gets the plugin in `requiredPluginIds`. - `npx vitest run src/__tests__/managed-instance-default.test.ts` (embedded Postgres): the earlier fill, lock and overlap cases pass. A sandbox fill also succeeds while active agents follow the default. - `npx vitest run src/services/managed-environments-plugin-entries.test.ts`: - The fill order is plugin then sandbox. - When the plugin is down there is no fill, and recovery never fills. - A failed ensure gives only the sandbox fill, and a failed fill keeps the ensure counted. - When the plugin fill declines, the sandbox still fills. - `npx vitest run src/__tests__/managed-config.test.ts src/services/managed-config-capabilities.test.ts`: parser refusals and the capabilities contract. - `npx vitest run src/services/managed-environments.test.ts src/__tests__/instance-settings-routes.test.ts src/__tests__/environment-routes.test.ts`: pass. The route still throws the same readiness conflict. - `pnpm --filter @paperclipai/server typecheck` passes. - I did not run the full `pnpm test:run`, `pnpm -r typecheck` or `pnpm build`. CI covers them. ## Risks - An instance whose default is unset at the first boot after this release gets the plugin environment, if its agents are ready for it. That covers a new instance, and an instance whose admin cleared the default or chose the local environment. This is the intended behaviour. - The readiness check doesn't lock agent rows. If an agent becomes ready during the boot fill, it can miss the check, which is the same race the admin route has. Agents created after the fill are prepared for the plugin, because setup reads the current default. - If the plugin is down at a new instance's first boot, the instance keeps the managed sandbox default from then on. - No schema change. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, through Claude Code with tool use (file edits, shell, test runs). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 16:48:13 -08:00
DottaandPaperclip	7ac99718a4	fix(workspaces): fence legacy sources for online migration (#19642 ) ## Thinking Path > - Paperclip coordinates AI agents and preserves their saved work. > - Remote workspace storage replaces legacy host directories with durable snapshots. > - Several legacy workspace records can name one physical project directory. > - A task lock alone cannot exclude a new task that selects that directory. > - Terminal recovery records can retain valid sources after physical writers settle. > - This change fences the source through its real preparation and lifecycle owners. > - Idle ordinary sources can then migrate while unrelated remote work continues. ## Linked Issues or Issue Description What happened? Online migration rejects ordinary legacy directories when several workspace records name the same path. A source inspection found 45 aliases across three project roots with no root Git marker or previously imported nested child. The instance also had unrelated running remote work. Requiring the whole instance to drain prevented useful migration of these idle sources. Terminal native sources and superseded finalization attempts also remain blocked after their exact physical and durable owners settle. Their files, provider allocations and recovery records must remain available through the change of file authority. Expected behavior Migrate an idle ordinary source under a derived physical source fence. Let unrelated object-backed turns continue. Preserve provider and recovery obligations. Keep unknown writers, shared Git stores and uncertain source state blocked. Steps to reproduce 1. Create several local workspace records for one managed project directory. 2. Give the aliases different source tasks. Leave their local writers idle. 3. Keep a valid remote workspace turn running on another source. 4. Inspect or apply migration to an alias. The old owner returns `overlapping_workspace_requires_maintenance`. 5. Retain a terminal native prepared source with a null coordinator owner, or an old finalize attempt with an exact later committed receipt. The old online path keeps these sources blocked. Paperclip version or commit Current tested head: `7478c817f09f7eed396922c0f7831af07ae6ff14`, based on master `8621bcf53ee2e4bbefb7d193add42dd544ca06e8`. Related recovery work: #19131. Deployment mode Hosted native Runner with remote workspace storage. Qualification uses disposable local PostgreSQL databases, actual archive/storage owners and synthetic source trees. The source observations were read-only. ## What Changed - Retain the existing company/task source keys and add one physical managed-project key. Heartbeat source selection, materialization, reopen and lifecycle owners share it. Configured aliases of canonical task directories retain their original task key. - Fence runtime provision and each configured service CWD through both startup owners. Check all unsettled service CWDs for physical source overlap, even when workspace IDs differ. Runs with no services keep their filesystem-free path. - Derive a bounded ordinary alias closure. Lock every alias task and lifecycle key. Reject foreign companies, changed membership, unproven descendants, included Git metadata and other unresolved writer obligations. - Commit remote authority on the connection that holds source exclusion. Refresh a stale selected local workspace after admission if migration published remote authority. Publish activity after commit. - Admit exact terminal native recovery sources under the existing physical finalizer keys. Bind descriptors, allocations and owner checkpoints to journaled fingerprints. Recheck them before capture and publication. - Add a separate company-scoped online entry for exact superseded finalization receipts. Keep the offline entry's deployment prerequisites and keep inventory read-only. - Document the proof and the remaining Git/common-store extension. The original files stay in place. No new scheduler, deployment action or database migration is added. ## Verification - Review found that the real reopen path could hold lifecycle exclusion while waiting for a project source. The new actual nested-reopen regression failed before the repair: lifecycle lock count was 1 while another connection held the project fence. The repair acquires declared and physical project sources before reconstruction and lifecycle. The contention case, ordinary-fetch restore and runtime nested restore passed. The lifecycle-worker restore first completed in 16.64 seconds but exceeded its unchanged 15-second body limit; a bounded warm rerun passed at that same limit. Both wrappers exited and completed cleanup. The source type graph, module-boundary check and diff check pass. Final 7478 CI passed: 58 successful jobs, three skipped, one neutral and a successful Snyk context. Fresh Greptile is 5/5 for exact7478, with no unresolved threads. [Final CI](https://github.com/paperclipai/paperclip/actions/runs/42318150314/job/128240631573). - Full CI on `e4283b3452` passed the other server, build and type checks. One exposure owner case failed: two execution workspaces received one preview pair after source admission serialized their starts. The first local seven-case run reproduced that failure; adding the owner argument alone did not change the hash payload. The completed repair binds run-scoped reuse to its execution-workspace owner. All seven existing real-process/fake-broker cases then passed in 13.81 seconds, with wrapper exit 0 and cleanup complete. Public scope IDs and explicit project/agent sharing are unchanged. The final7478 head passed all required CI and fresh review gates. - The service/source/restore owner canary passed 25 cases across four suites before the final no-service fast-path guard (41.33 seconds). This includes actual project service processes with a different project ID and no execution-workspace ID. Their startup waits for source ownership, migration stays blocked while they run, and the source becomes eligible after their verified stop. The final three-case runtime canary passed at the final source inputs (18.70 seconds). It covers that fast path and both service-CWD variants. Both wrappers exited 0 and completed cleanup. - The boundary-repair canary passed 9 cases across four suites at `7c18f1c170`. It covers the real rehome/conversion gateway, project command provisioning, future task selection, stale imported authority, foreign-company binding, and existing standalone filesystem callers. The shared PostgreSQL wrapper exited 0 and completed cleanup. Duration: 94.15 seconds. - The P1 diagnostic passed 46 cases across five suites at the pre-rebase P1 revision. It covers the actual importer with an export warning created by `recordNativeFinalizationFailure`, strict prepared-source/action bindings, finalizer locks, source/lifecycle races, unrelated remote execution, and bare Git with external alternates. It uses the repository shared PostgreSQL wrapper, `--fsModuleCache`, and a local-only 90-second hook allowance. Duration: 245.44 seconds. No product timeout changed. - The canonical retained-directory compatibility check passed 22 cases across two suites. Missing archived task directories still import from authenticated restore points; they do not need a present-directory source proof. - Both initial P1 regressions were observed before the fix. The real finalizer warning made the importer ineligible. The old archive owner accepted a nested bare repository borrowing external objects. The new cases pass and preserve the original warning, prepared descriptor, allocations and files. - Prior attempts remain non-passes: the first 23-case canary passed on an earlier head; its exact-head rerun hit 30-second database setup/cleanup limits. A read of the owned PostgreSQL server found a template clone waiting on `DataFileExtend` for 58 seconds. A broader owner run was intentionally interrupted during slow existing heartbeat setup. One diagnostic invocation used an invalid CLI option and executed no tests. A later review canary passed 21 cases and failed the new real restore contention case: cwd-only admission missed the declared project of an external worktree. Its fixture also omitted the required file-repository authorization callback and produced an unhandled rejection after the failed assertion. The repaired run passed 22 cases, exited 0 and completed cleanup. These earlier failures are not reported as passes. - The recovery owner passed 37 PostgreSQL cases, including 14 new and 23 existing finalized-recovery cases. The operation reconciliation owner passed 40 cases. The initial archive run passed 103 cases and found a missing closed diagnostic entry; the corrected diagnostic and expanded Git rejection cases pass in the P1 run. - The full server type graph passes with the Runner source exports mapped locally because this checkout has no built Runner declarations. The latest module-boundary check and `git diff --check` pass. Standard package build/typecheck and broad final-head regression coverage remain required in CI. - The initial CI exposed missing admission scopes for standalone filesystem APIs and a changed foreign-company error. The latest repair keeps those APIs independent of a database and holds physical fences at their actual database-backed mutation owners. The branch also includes the existing master fix for unchanged Runner default-model assertions. New review findings and new-head CI are being checked. - The local canary command and bounded source selection are documented in `doc/plans/2026-11-06-online-ordinary-source-fence.md`. Source observations only read data. A fresh bounded metadata probe found 254 pending legacy records and zero unsettled local-process services. No live import, hold, cancellation, provider action or source filesystem write was performed for this change. ## Risks - Run-scoped reuse now includes execution-workspace ownership in its private key. Explicit project/agent sharing keeps its previous identity. - A missing writer owner could make a source proof unsafe. Admission derives physical aliases, includes future project selection and retains all existing source/lifecycle/finalizer locks. Unknown and changed state fail closed. - Cross-company physical alias checks use bounded minimal-path pages. Lexical-prefix filtering cannot safely replace them because a symlink can hide a foreign alias. Unknown paths or more than 10,240 foreign records leave the source explicitly unproven. More than 1,000 same-company local records also fail closed. - The project key can serialize local preparation within one project. A first project waits; additional project acquisition fails on contention to avoid a lock cycle. Remote authority keeps its remote preparation path. - Recovery preservation does not prove provider termination or successful file recovery. Existing provider allocations, checkpoints, task intent and recovery actions remain authoritative. - This slice does not finish shared Git migration. Git administrative/common-object/alternates closure and exact nested imported-child coverage remain the next owned extension. No global writers-stopped shortcut is used online. ## Model Used OpenAI GPT-6 through Codex, with reasoning, tool use and code execution. The host does not expose a more specific backend model ID or context-window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 18:42:45 -06:00
Devin FoleyandPaperclip	2a4a001d37	Preserve GitHub access repair guidance for MCP callers (#19655 ) Return GitHub permission denials as failed MCP tool results with trusted, connection-specific recovery guidance. Explain the personal-account and organization App installation boundary without replaying writes or bypassing provider-content checks. Align the repair steps with the Permissions UI. Cover enabled model checks and malicious provider failures across direct calls, connection tests and stored approvals. Repair the CI-exposed synthetic onboarding environment fixture without weakening cancellation assertions. Verified 311 targeted tests, six onboarding Playwright tests, server typecheck, module boundaries, full CI and Apex 5/5 with all threads resolved. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-11-06 16:32:27 -08:00
DottaandPaperclip	a3eecf1fc9	Preserve admitted runtime skill bundles during server replacement (#19675 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Native runs keep the exact skill files admitted for execution and recovery. > - Older inputs can name immutable local bundles without a skill version ID. > - Installed skill heads alone do not preserve those admitted bundles. > - Server replacement needs a durable source for each retained bundle. > - This pull request copies authenticated bundles through the existing package owner and restores their exact projections. > - The benefit is preservation during online migration without stopping active work or settling recovery warnings. ## Linked Issues or Issue Description Refs #19656. Related: #19660. What happened? Older native execution inputs can contain skill bundles with a null version ID. Recovery skips those bundles. Startup installation checks can pass even when those retained inputs still require the old filesystem. Expected behavior Preserve each retained admitted bundle in object storage. Restore its original bytes, modes and digests on an empty replacement host. Keep its null pin and all run, controller and recovery state. Block startup and enrollment while a concrete local asset obligation remains. Steps to reproduce Retain a native run with an admitted local skill bundle and a null version ID. Convert its installed skill head to durable storage. Replace the server home and attempt recovery. The old bundle path remains in the recovered context. ## What Changed - Add a bounded `runtime-assets` phase to the existing skill package maintenance CLI. It copies exact admitted files under the runtime asset publication lock, verifies the durable package, and appends a run-bound receipt after source revalidation. - Keep original files, null pins, task intent, warnings and controller leases. Support an active run through copy-only preservation. Do not start a turn or claim provider ownership. - Verify and reuse a healthy canonical local bundle before storage reads. Restore missing or damaged bundles through the existing package and native asset materializers. Use the authenticated original manifest to preserve exact read and execute modes. Accept a later contract revision only when the run binding and exact skill selection remain unchanged. - Check retained recovery obligations and current task sessions during both startup and migration eligibility. Ignore ordinary admission that has no completed execution input. - Protect receipt-backed packages from orphan collection. Serialize receipt publication and deletion on the package row, then check references in a fresh statement after the lock. - Add real PostgreSQL and filesystem regressions. Replace a delete-method proxy with an actual upload-renewal row-lock race. Publish mutation activity only after commit. ## Verification - Baseline production startup gate failed the new concrete-asset regression as expected. The fixed gate passed. The actual before-fix preservation owner also failed recovery from a healthy local bundle during a simulated storage outage. A verified local-first path fixes that regression. - Eight preservation tests passed with real PostgreSQL, streaming storage, archive extraction and native materialization. They cover cold recovery, exact modes and binary bytes, active ownership, changed sources and bindings, contract revisions, missing or corrupt receipts, and publication/collection races. - Six affected package collection/deletion controls and three existing startup controls passed. The existing versioned recovery consumer passed a sequential retry. - One fixture attempt failed on its cleanup foreign-key order; the fixture was corrected. One existing suite hit its unchanged 30-second database startup hook before test execution. The unchanged single consumer passed on retry. These failures remain recorded. - Source, changed-test and CLI TypeScript checks passed. Rebase preserved both the local instruction-copy and runtime-asset guards. Current-head changed-test checks passed. Final PostgreSQL rerun passed all eleven preservation/startup cases. Current CLI check passed. Required CI remains pending at the trusted-runner gate. - Current canonical bundled creation uses package preparation plus atomic content/head publication. Native preparation requests immutable version snapshots before delivery. Converted installations and new canonical bundled creation therefore do not recreate the historical null-version asset dependency. - Fresh Greptile review on exact head `54c7dd048c9016c1af9fc68213d5ae053ac16499` passed at 5/5 with zero unresolved threads. - Module boundaries, whitespace, and credential/internal-reference scans passed. No live source writes, provider actions or migration apply were performed. ## Risks Missing, changed or unsafe local files remain blockers. This command does not repair a previously published package whose objects later disappear or become corrupt. It also refuses malformed preservation receipts. It retains the original local files and prior evidence and keeps removal blocked. Automatic repair requires a separate canonical package-owner recovery operation; direct database edits are not an accepted recovery path. Failed publication keeps the original files and any unreferenced package for normal collection. The readiness check is explicit startup or enrollment work, not a background scan. Its cost grows with retained admitted asset obligations. Older readers must be upgraded before the old volume is removed. This change preserves asset bytes; it does not settle business recovery or qualify unrelated volume dependencies. ## Model Used OpenAI GPT-6 (Codex). The runtime did not expose a more specific backend model ID or context-window size. Used reasoning, code editing, and tool execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 18:23:53 -06:00
DottaandClaude Opus 5.5	6d1ecfc420	fix(accounting): close a remote run's lost receipt once its sandboxes are terminated (#19606 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Budget admission waits while completed runs await accounting; #19497 made that a hold on new work, never a hard stop > - #19497 also closes a lost usage receipt once a local provider is verified stopped, so the hold ends > - A remote (Daytona) run has no local stop event and keeps its recorded remote process identity, so its lost receipt never closed. The company then held all new work until a receipt that can no longer arrive > - This pull request closes a remote run's lost receipt once every sandbox it used carries a provider termination receipt > - The benefit is that a restart that kills a Daytona turn no longer stops a company's new work ## Linked Issues or Issue Description Follow-up to #19497. Refs #19544 (Daytona chaos harness). What happened? In a Daytona restart wave, run `a261c30b` failed after native recovery was blocked. Its usage receipt stayed incomplete (`accountingReceiptReady: false`). Both of its sandbox leases were released with a `destroyed` termination receipt, but accounting stayed pending. Admission then refused new company work with "New work must wait while a native run recovers its unfinished accounting". What did you expect? Once every sandbox of the run is stopped or destroyed, no runner can report usage for it. The known part of its usage should settle as an unpriced charge, as it does for a verified local stop, and new work should be admitted. Steps to reproduce 1. Run a native task on a Daytona environment with a hard-stop budget. 2. Kill the control plane during the provider turn so the turn's receipt is lost and native recovery ends in terminal failure. 3. After the sandbox lease is released with a termination receipt, observe that the run stays `cost_accounting_pending` and new work is held. ## What Changed - `run-cost-accounting.ts`: the closed-receipt predicate gains a remote branch. A native run in terminal failure without a result (failed, cancelled, timed out or interrupted) is closed when it has at least one environment lease and every lease carries a `paperclip.remote-termination.v1` receipt bound to that lease, run, company, provider and provider lease ID, with state `stopped` or `destroyed`, after a successful release. This mirrors `hasRemoteTerminationReceipt`. One SQL predicate still serves the reconciler scan and settlement. - The local branch is unchanged: a failed run with a verified local stop event and a cleared process identity. - A sandbox that may still run (no receipt, lease still active), a receipt bound to another sandbox, or a run with no lease keeps its debt pending. - `budget-enforcement-worker.ts`: the accounting worker also wakes when environment-cleanup or retained-session-cleanup work commits. Those are the signals the lease writers already send when they release or stop a sandbox. - `doc/SPEC-implementation.md`: documents the remote closure rule. ## Verification - `closed-native-receipt-accounting.test.ts` (embedded Postgres): - A failed and a cancelled remote run, with one destroyed and one stopped sandbox, settle as one unpriced charge, settle the reservation, and clear admission. These are red on master. - Debt stays pending for an active sandbox with no receipt, for one of two sandboxes without a receipt, for a receipt bound to another sandbox, and for a run with no lease. - `budget-enforcement-worker.test.ts`: the accounting worker wakes on environment-cleanup and retained-session-cleanup commits, and stops waking after `stop()`. These are red on master. - Existing suites pass locally: native-stop-accounting, budget-enforcement-scoping, heartbeat-cost-accounting, run-cost-accounting, cost-data-quality, budgets-service. Server typecheck passes. - Live Daytona restart wave: results in the comments. ## Risks - A cancelled remote run now closes once its sandboxes are terminated. Stop recovery could otherwise reattach a recorder, but a stopped or destroyed sandbox has no runner to reattach to. Local cancelled runs keep their drain. - Workspace export can restart a stopped sandbox and removes its termination receipt first. A receipt that arrives after settlement still posts through late-receipt reconciliation. - Reusable-lease chains (`resolveRemoteTerminationLease`) and recovered-cleanup stop receipts are not mirrored, so those runs stay pending. That is the conservative side. - Low query cost: the predicate runs only for pending terminal runs. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file editing, embedded Postgres tests, live restart-chaos harness). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 18:17:11 -06:00
DottaandPaperclip	b8d94b3ad1	test(server): verify status and object deadlines across cold sleep (#19347 ) ## Thinking Path > - Paperclip manages agents and their useful work. > - Status cards and external objects can have future refresh deadlines. > - A stopped tenant needs a durable Cloud alarm to return for those deadlines. > - A local quiet report alone does not prove that handoff or later useful work. > - Acceptance must exercise the normal producer, authorization, claims and alarm owner. > - This PR adds opt-in signed cold-owner acceptance and bounded launcher controls. ## Linked Issues or Issue Description Refs #19001 and #19191, which have merged. This PR targets master and adds acceptance for the existing status and object owners. What happened? The existing local coverage did not join useful status generation or built-in GitHub refresh with signed Cloud alarm persistence, physical child retirement and cold reactivation at the original deadline. Expected behavior Keep the feature enabled. Complete useful work through its normal owner. Persist the exact authorized alarm before stopping the child. Wake and process the same original deadline without losing identity, claims, retries or useful results. Steps to reproduce Use the checked-in Node24 launcher with separate owned loopback App/Cloud test databases, a compiled Cloud checkout and the pinned Node22 executable. Select one status or object bootstrap, cycle or identity case. The command validates fresh evidence and database ownership before starting its guarded worker. Paperclip version or commit Current candidate `ea911278b22c855d6722e43afc026bfc5c031405` normally merges fetched master `07a548073ab48f13cb93b2fdc31553a5da88a070`. All eight owned acceptance files are byte-identical to `c08e64fa857d39bdb79352e94d156db7a4dbebf3`; all unrelated paths exactly match master. The guard-argument repair and source-specific historical cycle proof are preserved. ## What Changed - Add one named opt-in case per owner/stage/diagnostic combination. Keep the shared business body and its original deadlines, assertions and cleanup. - Drive actual managed App startup, authenticated producers, persisted claims and the signed Cloud SQL alarm before physical child stop. - Retain the generated five-secret fixture profile across cold boots. Store it in private directories and retain only names, hashes and modes in evidence. - Add optional query timestamps to the existing proxy and harness. Existing callers keep their default behavior. - Add launcher refusal controls and assert its complete child argument list. Both explicit guard preloads and the exact child `NODE_OPTIONS` must match. ## Verification - The current preflight repair has causal proof: the old controls pass all nine cases with an invalid worker preload flag. The new assertions reject that same command in both success controls (two failures/seven passes). Restoring the correct command passes all nine cases. The original refusal, freshness, symlink and byte-preservation assertions remain unchanged. All 33 recorded Node PIDs and three groups close; temporary copies and private directories are removed. These controls stop at the declared spawn stub. They verify the full command contract and do not claim to execute Vitest. - The prior real named status bootstrap passed in 72.942 seconds: one selected case/six explicit skips. It exercised the actual launcher and Vitest fork, useful normal generation, current authorization, settled claims, original 900000ms interval and signed SQL alarm generation/operation-lock acknowledgement before physical child stop. Recorded processes and final SQL sessions closed; both successful databases were normally dropped. - Earlier normal status and GitHub-object cycles qualified their real 900-second and 300-second original deadlines, cold reactivation and useful second results. Their complete original failure and recovery records remain source-specific; no shortened business clock or manufactured completion is used. - On current integrated head `ea911278`, own frozen offline setup and normal shared/SDK, Runner and App declaration prerequisites passed. The status test and its imported closure pass the affected normal test-tree compiler program with original options/references. The actual named normal status bootstrap then passed: one selected case/six skips, 82.202 seconds in Vitest and 85.616 seconds in the launcher. It used the real checked-in launcher, normal App/SDK/process agent and exact compiled Cloud `389878364b42dd129835d263f91882c5774d5123`. The original 900000ms interval and 120000ms alarm lead remain unchanged; useful v1 and signed SQL alarm generation/operation-lock acknowledgement precede physical App stop. All six recorded process IDs and the runtime group were absent; independent final test-database sessions were zero, with no observer during execution. Both fresh success databases were normally dropped. This refreshes the changed consumer seams through handoff and stop; it is not another cold-wake/useful-v2/full-cycle result. Historical cycles remain tied to their measured source. - Fresh exact-head CI and complete edited bot reviews are required before merge. A whole final-tree local build/test or deployed suspension is not claimed. Local provider, process and S3 boundaries are controlled fixtures; no live GitHub provider or Railway/Neon suspension claim follows. ## Risks - These tests use real SQL and wall clocks and are opt-in. Their fixtures must stay isolated and retain failed evidence. - Stable fixture secrets must remain private and must not appear in logs. Cleanup keeps them when owner closure is uncertain. - The acceptance body stays shared to preserve measured evidence. The named registration and command controls make selection failures visible. - This PR does not change production, schema or feature enablement. ## Model Used OpenAI `gpt-6.1-sol` produced the delegated acceptance work. OpenAI Codex, based on GPT-6, performed managing review, source integration and the preflight argument repair. Both used repository tools and code execution. The coordinator's exact backend snapshot and context-window size were not exposed. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used - [x] I have checked ROADMAP.md and confirmed this does not duplicate planned core work - [x] I have searched related PRs and linked the existing foundations - [x] I have described the issue with the bug-template fields - [x] I have not referenced internal or instance-local issues or links - [x] My branch name describes the change - [x] I have run the stated focused local tests and preserved their limits - [x] I have added or updated tests - [x] I have documented the opt-in command and evidence requirements - [x] I have documented risks - [ ] All current-head CI gates are green - [ ] Current-head Greptile is 5/5 with no open findings - [x] I will address reviewer findings before merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 18:15:56 -06:00
DottaandClaude Opus 5.5	7c673e695f	fix(chat): name tasks by identifier, not UUID, in tool rows and agent text (#19635 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - New users meet Paperclip through the onboarding chat with their first agent, and through task briefs that agents write > - A browser walk of onboarding found internal handles and control-plane jargon in those user-facing surfaces > - A tool row showed `Read taskId: e3ecf9bc-…`, an agent-written brief cited `task … (id 2b330993-…)`, and a Codex agent told a new user about "receipts", "execution policy", and a "rejected policy patch" > - Raw UUIDs and mechanism narration make the product look broken and give the user nothing to act on > - This pull request makes tool rows name a task by its identifier, and tells agents to reference identifiers and to describe outcomes, not mechanics > - The benefit is that a new user sees "Read <task identifier>" and plain progress updates ## Linked Issues or Issue Description What happened? In the task chat, a collapsed tool row rendered the raw call argument `taskId: <uuid>`. An agent-written task description included a raw task UUID next to the task identifier. A Codex chief of staff narrated internal mechanics (receipts, execution policies, a rejected policy patch) to a new user in chat. Expected behavior Tool rows name the task by its identifier. Agent-written text links tasks and documents by identifier. Agents tell the user what happened, what it means, and the next step. Steps to reproduce 1. Run onboarding with the Codex API runner and accept a single-task proposal. 2. Watch the handoff in the chief-of-staff chat. 3. See the task lookup tool row and the narration. With the Claude API runner and the plan path, open a child task created from the plan and read its description. ## What Changed - `ui/src/components/task-chat/transcript-adapter.ts`: a tool row's target never contains a UUID argument. A `taskId`/`issueId` UUID is kept as `taskRefId` on the tool item. - New `ui/src/components/task-chat/task-reference.ts`: `useTaskReferenceLabel` reads the task from the shared issue detail query cache and returns its identifier. It returns "task" until the task resolves or when it cannot. - `TaskChatRunnerActivityGroup` and `TaskChatToolCard` show that identifier as the row subject. They load the task only for rows that have a task reference. - `skills/paperclip/SKILL.md` (Comment Style): two new required rules. "No raw IDs in prose" and "Describe outcomes, not mechanics". - First-task onboarding skill: a "Talk in outcomes" rule. Non-inline handoff inputs use an identifier link, not a raw UUID. - Chief-of-staff persona (`AGENTS.md` for the first agent): the same plain-language rule, so it applies in every chat. - `skills/paperclip-converting-plans-to-tasks/SKILL.md`: task descriptions reference other tasks and documents by identifier link, never by raw UUID. - No change to `skills-releases/` snapshots. They are frozen releases. ## Verification Red on master (fix files reverted, new tests kept): ``` ui tool-row-task-reference.test.tsx FAIL ... > falls back to a plain word while the task cannot be resolved AssertionError: expected 'ReadtaskId: e3ecf9bc-099a-4b10-8169-9…' not to contain 'e3ecf9bc-099a-4b10-8169-9ce490a91e3d' Tests 3 failed (3) server user-facing-wording-guidance.test.ts Tests 6 failed (6) ``` Green with the fix: ``` ui tool-row-task-reference.test.tsx Tests 3 passed (3) server user-facing-wording-guidance.test.ts Tests 6 passed (6) ``` Neighbouring suites: - `cd ui && npx vitest run src/components/task-chat src/components/TaskChatThread src/components/transcript`: 70 files, 1448 passed, 1 skipped. - From the repo root: `onboarding-first-task-assets`, `paperclip-skill-utils`, `onboarding-first-task-admission`, `hiring-operational-examples`, `cli-invocation-safety`, and the new guidance test all pass. - `pnpm check:token-gates`: all gates clean. The change adds no styles. - `ui` typecheck: no errors in the changed files. - Not run locally: `agent-skills-routes.test.ts` hit a 30 s `beforeAll` import timeout on a cold cache. CI covers it. ## Risks - Low risk. A row that names a task loads that task once through the shared issue detail cache. The thread usually has it cached already. A failed load shows "task" and does not retry. - Rows now show other non-UUID arguments after the identifier (for example `<identifier> · status: done`). Rows without a task reference do not change. - The skill text changes the guidance that agents read. They do not change permissions, tools, or API behavior. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, run in Claude Code with tool use (shell, file edits, test execution). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 18:12:48 -06:00
DottaandClaude Opus 5.5	6c7cc2c5d6	fix(native): do not spend a recovery attempt resuming a cleanly drained run (#19625 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Native runs have an execution recovery budget of three provider attempts, so a run that keeps failing stops retrying > - A graceful shutdown (deploy or SIGTERM) drains native runs: it suspends them for reattachment by the next server > - When the runner did not survive, the next boot resumed the turn and charged one attempt > - Restarts that the run did not cause used up its budget, so recovery stopped and the task needed a manual Retry > - This pull request does not charge an attempt for resuming a run that a graceful drain suspended > - The benefit is that deploys and graceful restarts no longer exhaust a healthy run's recovery budget ## Linked Issues or Issue Description What happened? In restart-chaos testing, native runs reached `execution_recovery_budget_exhausted` (3/3) only because of server restarts. In one wave, two restarts crossed each of two runs. In a second schedule (K1, then K2 when ready, then K2 again), each drain correctly suspended the local ACPX Claude run, yet the fourth boot blocked it with `native_restart_recovery_blocked`. Every resume after a drain was classified `resume_dead_runner` and charged an attempt (`nextNativeProviderAttempt` in `native-restart-recovery.ts`, applied by the executor's execution claim). Expected behavior A restart that the run did not cause should not consume its recovery budget. A graceful drain records `restart_suspended` and `awaiting_runner_reattach`. The resume that follows should keep the attempt count. Paperclip version or commit `master` at `4ef3fa4271`. Steps to reproduce Start a native local run. Send the server SIGTERM three times while the run is active, with a full restart each time. The third resume is blocked with `execution_recovery_budget_exhausted`. ## What Changed - `suspendedByCleanDrain()` in `native-restart-recovery.ts` recognizes a run whose most recent recovery claim is older than the drain's `restart_suspended` entry. Ownership waits and deferred blocks between the drain and the next claim do not erase it. The next claim consumes it. - A scan that waits for the old controller to exit (`awaiting_evidence`) records one history entry per reason, so repeated scans cannot push the drain record out of the bounded history. - Restart claims carry `chargesProviderAttempt`. It is `false` for a claim that follows a clean drain. - `nextNativeProviderAttempt()` takes that flag. The executor passes it when it takes the execution lease, so a drained resume keeps the current attempt. - The first execution is always counted: a run drained at attempt 0 (no execution claimed it yet) still charges its first provider start. - A run drained during its third attempt may continue that attempt. The local resume and remote bootstrap gates skip the `>= 3` exhausted check only for a non-charging resume, and the executor still refuses a fourth attempt. - A crash restart (no drain record) still charges one attempt. Reattachment to a live runner still charges none. - `doc/execution-semantics.md` states the rule next to the three-attempt budget. ## Verification - New test in `server/src/__tests__/heartbeat-native-restart-suspension.test.ts` (embedded Postgres, real heartbeat drain and restart recovery): - Red before the fix: the claim after the drain had no `chargesProviderAttempt: false`. - Green after the fix, for drained attempts 0, 1 and 3: the crash claim charges (`true`). After the drain, a controlled backend lets the executor take its execution lease and records the saved `native_run_finalizations.attempt` at provider dispatch: 1, 1 and 3. Attempt 3 is claimed instead of blocked. - Red at the executor: when the executor ignores the flag, the drained attempt-1 resume saves 2. - Ownership-wait case: the drained controller (the test process) is still alive, so two scans wait. The history keeps `restart_suspended` followed by one `awaiting_evidence` entry. After the controller is gone, the claim does not charge, and attempt 3 continues. - Classifier cases (local resume and remote bootstrap): attempt 3 after a crash is still `execution_recovery_budget_exhausted`. Both new cases were red on the previous revision. - `heartbeat-native-restart-suspension.test.ts` 12/12, `native-runner-restart-recovery.integration.test.ts` and `native-runner-ownership.test.ts` 94 passed. - `tsc --noEmit` for `server` passes. ## Risks - A run that itself makes the server shut down gracefully, again and again, would no longer be stopped by this budget. Graceful shutdowns are operator or deploy actions, so this is unlikely. A crash loop (SIGKILL, OOM) still charges an attempt on every restart, and a resumed turn that fails still goes through the existing failure and replacement budget. - This changes only the attempt count. Ownership checks, checkpoint verification and the exhausted gate are unchanged. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file edits, test execution). Extended reasoning enabled. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 18:12:12 -06:00
Nicky LeachandClaude Opus 5.5	55244f1e9c	test(server): remove two timing flakes in cloud adapter and native cleanup tests (#19668 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Contributors wait for PR CI before merge, and a flaky test forces a re-run of the whole CI run > - Four server tests failed on 2026-11-06 on commits that passed on a later attempt > - Two of them failed because of real slowness inside a fixed time limit on contended 4-vCPU runners > - This pull request removes that slowness from the two tests. It does not raise their time limits > - Of the other two tests, one is already fixed on master, and one is not reproduced (see below) > - The benefit is fewer whole-run re-runs ## Linked Issues or Issue Description What happened? These tests failed in PR CI and passed on a re-run of the same commit: - `server/src/adapters/cloud-external-adapters.test.ts` › "Cloud image adapter boundary › cold boots built-ins…" failed with `Test timed out in 15000ms` (3 times). - `server/src/__tests__/retained-native-session-cleanup-work.test.ts` › "drains a backlog across discovery pages…" failed with `Test timed out in 15000ms` (2 times). Expected behavior Both tests pass on a loaded CI runner. Steps to reproduce 1. Run the test file in a 4-CPU container (`docker run --cpus=4 --memory=16g`) with 4 copies at the same time. 2. Before this change, the adapter case fails in 4 of 4 runs at 15,008 to 15,041 ms. ## What Changed - `cloud-external-adapters.test.ts`: add a `beforeAll` warm-up with a 90-second limit. - The warm-up imports `registry.js`, `plugin-loader.js` and `adapter-plugin-store.js` with an empty temporary home and the managed signal, so it reads and runs no persisted adapter package. - Before this change, the first case paid the one-time transform of the whole built-in adapter graph inside its own 15-second budget. In CI the first case timed out at 15 seconds, and the next case took 1.1 seconds. - Each case still gets fresh module state from the existing `vi.resetModules()`. No assertions change. - `retained-native-session-cleanup-work.test.ts`: - `flush()` takes an optional poll `interval`, which defaults to 50 ms, so other tests do not change. - The backlog test uses 5 ms. The worker admits one cleanup pass per 1-second fake-time step, so 103 entries need 103 polls in a row. - The next wake is scheduled before a pass releases its idle-work hold, so a faster poll cannot see a false quiet state. No assertions change. ## Verification \| Test \| Before \| After \| \| --- \| --- \| --- \| \| Adapter cold boot, container, 4 copies \| failed 4/4 (15,008–15,041 ms) \| passed 4/4 (537–756 ms per case; the warm-up takes about 23 s) \| \| Adapter cold boot, host \| failed 2/2 \| passed 3/3 (0.6–1.6 s) \| \| Backlog drain, container, 4 copies with load \| 6.3–6.6 s \| 2.0–2.1 s \| \| Backlog drain, host, unloaded \| 5.1 s \| 1.1–1.3 s (5 runs, 29/29 pass) \| - Final host run of both files with `--maxWorkers=2`: 33/33 pass. - The backlog test did not reach the 15-second timeout in local runs. The evidence for that fix is the measured poll-bound cost. Not changed in this PR: - `connection-management.integration.test.ts` (expected 3 to be 2): - Cause: a GitHub webhook recovery worker from an earlier test fetched through the suite-wide fetch spy 60 seconds later. - Master already fixes this with commit `0b48bc7ac5` (#19541). The failed jobs ran on bases without that commit. - `idle-completed-history.test.ts` (runtime service never ready): - Not reproduced in a 4-CPU container, even with heavy load. CPU load alone does not explain a child process that stays silent for 10 seconds. - A next step is to record spawn-to-listen timing in CI. ## Risks - The adapter warm-up adds about 1 to 25 seconds to the file's `beforeAll`, depending on load, but the test body becomes faster by the same amount. - If the adapter graph import becomes much slower, the 90-second warm-up limit shows that as a clear hook timeout. - Typecheck did not run locally. CI checks it. The edits are a `beforeAll` import and a numeric default parameter. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`) in Claude Code, with tool use and code execution. A Claude sub-agent found the causes and made the fix, and the parent session reviewed it. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 16:02:56 -08:00
DottaandPaperclip	6be71028ad	Prepare online ephemeral enrollment through signed Cloud control (#19660 ) ## Thinking Path > - Paperclip manages agent work through a control-plane web server. > - A replacement web server needs shared durable data and the original signing identity. > - Existing enrollment exports that identity only after a global drain and offline migration. > - Online enrollment needs a separate preparation operation while agent work continues. > - This change adds a purpose-bound Cloud control operation with existing data and key owners. > - The result is an audited preparation receipt, with no permission to stop agents or detach storage. ## Linked Issues or Issue Description Refs #19038. Refs #19024. What are you trying to improve? Allow Cloud to prepare an already migrated tenant for volume removal without the offline migrator or a global-idle gate. What problem does this solve? Cloud must preserve the original decision key before starting a replacement with static ephemeral mode. A tenant session must never gain key-export authority. What would the improvement look like? A fresh exact-action Cloud assertion obtains a private, deployment-bound preparation receipt. Unfinished durable-data migration and invalid replacement configuration block export. ## What Changed - Add the signed `ephemeral-enrollment:prepare` control action and a strict UUID-only POST route. - Check existing durable-data eligibility and ephemeral configuration before reading the original key. Never create a missing key. - Audit only public preparation metadata. Keep HTTP diagnostics secret-sensitive and the response non-cacheable. - Bind the receipt to the attempt, stack, build commit, Railway deployment, time, static mode and key digest. - Always return `enrollmentReady: false`. Cloud still needs to arm a replacement, verify old writers have stopped, repeat eligibility and qualify native handoff before detaching the retained volume. ## Verification - PASS: 80 focused tests across the preparation owner, signed route, existing Cloud control, HTTP privacy and OpenAPI route coverage. - PASS: 2 real PostgreSQL cases. Concurrent copies of one signed request export once; the durable receipt still rejects reuse after clearing the process fence. Normal run admission before skill pins is allowed; missing durable skill installation remains blocked. - Genuine prepatch RED: the same two PostgreSQL cases exported twice and rejected ordinary admission. The fixed owner was restored automatically after the comparison. - PASS: server TypeScript check with workspace dependencies built. - Fixed the first CI route-coverage failure by declaring this private control endpoint in the existing explicit OpenAPI exclusion registry. - Initial local HTTP execution lacked sandbox socket permission; the reviewed run passed. Initial PostgreSQL fixtures enabled durable files too late and reused an issue prefix; those fixture defects were corrected before the red/green comparison. - All required CI gates pass on `5346721847`. Fresh Greptile review is 5/5 and all review threads are resolved. No deployment, volume detachment, admission change or agent stop is performed by this PR. ## Risks - The private response contains the original signing key. Only a fresh stack-bound assertion for this exact action can obtain it. The Cloud consumer must pass it directly to its existing vault owner and must not expose it through admin responses or registry metadata. - Eligibility is an observation. It cannot replace static writer restrictions or qualified ownership handoff. This PR deliberately supplies no detach or readiness authorization. - Cloud transition integration is still required. This PR alone does not enroll an instance. ## Model Used OpenAI Codex, GPT-6 family, with reasoning, code execution and browser tools. The host does not expose the exact serving model ID or context-window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 18:00:28 -06:00
Dotta Paperclip Codex	c82211251b	test(chat): cover Slack forwarding deadlines through registered owners (#19348 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - People answer agent questions through Slack commands, buttons, and forms. > - Slack requires a response within three seconds. A modal trigger also expires after three seconds and can be used once. > - Durable event delivery does not prove these synchronous response paths. > - This pull request exercises the registered app owners and pinned SDK with original signed requests and controlled forwarding delays. > - The tests record deadline failures without clearing invalid forms, extending triggers, or changing production behavior. ## Linked Issues or Issue Description Refs #19019, which has merged. This PR now targets `master`. Related work: #8300 handles raw-body capture and plugin event fan-out. #19328 handles Slack connection onboarding. This PR changes neither path. I searched related open and closed work and read those PR bodies and changed paths. What happened? A forwarding delay can leave an authenticated Slack callback too late to acknowledge a command, exchange a modal trigger, or return field errors. The existing direct modal test did not cover the normal registered ingress or these provider deadlines. Expected behavior Tests must distinguish useful app persistence from a response that Slack can still accept. They must preserve current authorization, field validation, original signatures, and exactly-once effects. Steps to reproduce Use an owned PostgreSQL test database and the repository launcher below. Run warm first. The default run adds one real 3500ms delay independently before command, form-open action, and invalid submission forwarding. Paperclip version or commit Production remains byte-identical to `54fa701d41b4ced62f4ff338d39b946015670ae3`. This test-only candidate is `a3cce4e9868fd3e6e25d9df7d0090899bdf6a204`. It adds a recovery-fixture setup guard after `dac0010649c435f673172c496b3383e99780e8bf`. Deployment mode Source-checkout integration test with real PostgreSQL, registered app owners, and the pinned Slack SDK. Provider API responses and the heartbeat receipt helper are controlled fixtures. No Slack account, model process, deployment, or physical tenant suspension was tested. ## What Changed - Route original signed command, action, and modal requests through the normal chat service and real SDK runtime. - Check forged-signature refusal, callback evidence, current-role refusal, invalid-field errors, one corrected answer, and duplicate fencing. - Intercept the pinned SDK HTTP transport. Model trigger expiry and one-use exchange without changing request timestamps or signatures. - Observe the real platform exception through the actual bound `views.open` callable. Delegate once and rethrow the same error. - Restore the network guard after every setup failure. Attempt all owned cleanup and preserve the original error before any cleanup errors. - Guard auxiliary database cleanup when recovery test setup fails before it creates a client. Keep temporary-cluster cleanup reachable and preserve the original setup error. - Add a repository launcher with the current prewire guard, one worker, explicit test-database validation, and closed warm-only and action-delayed-only selectors. The default still selects all four cases. ## Verification - Current candidate `c6ae7957988e220ea096ab678f070d3bd1c5f9c0` normally merges fetched master `8621bcf53ee2e4bbefb7d193add42dd544ca06e8`. All three owned protocol/recovery acceptance files are byte-identical to the preceding head. All other paths exactly match master. The shared Runner default test conflict uses master's independently landed constant-based repair. That supersedes this branch's temporary two-literal repair; no fourth file remains in the PR diff. - Original CI server4/12 had two stale `gpt-5.6-sol` expectations after the normal default changed to `gpt-6.1-sol`. The complete failure log is retained. Before the master repair landed, the focused two-literal correction passed the whole six-case file with explicit-model, non-Codex, lifecycle, permission and skill checks intact. That local result is source-specific; it is not a fresh full merged-tree pass. - The original guard-leak counterfactual reproduced unrestored global network functions after setup rejection. The repaired callback passed four controlled cleanup diagnostics. Cleanup attempts each initialized owner and restoration even after a failure. It preserves the primary error first and retains separate cleanup errors. - A fresh registered SQL/SDK warm case passed on the qualified protocol source with one persisted answer, 17.15 seconds total and normal original assertions. Final independent SQL and recorded PID/group audits were empty. A diagnostic sampler initially hit EPERM, and the worker exit guard line was absent; those limits remain explicit. The cleanup diagnostics are local qualification tools, not four newly checked-in CI tests. - Delayed modes remain explicit diagnostic commands. Ordinary CI runs warm. Earlier mode results do not establish a final-source four-case single-command pass. The provider transport is controlled; no live Slack display, deployed sleep or wake result is claimed. - The original server6/12 cleanup-backlog case hit its 15-second whole-case timeout. Its phase and cause remain unknown. The original E2E shard received runner shutdown without an assertion verdict. Complete original logs are retained; neither failure has been labeled a flake or repaired by extending a deadline. - Own offline dependency setup passed with an unchanged lockfile. An omitted private guard dependency initially prevented test entry; that setup failure is retained separately from product results. The immutable helper closure corrected it. No local full-repository typecheck/build/test pass is claimed. Fresh exact-head CI and complete edited bot reviews are required before merge. ## Risks This is a test-only change. It changes no production code, credential flow, schema, API, or sleep eligibility. The existing large integration fixture remains large; this PR does not cross a file from below 1000 lines to above it and adds no code comments. The provider transport is synthetic. The tests demonstrate deadline consequences and app persistence, not that Slack displayed a late response. They do not make commands or open forms safe while the tenant sleeps. Other Slack lanes and persistent transports remain open work. CI may expose additional harness assumptions. The roadmap's Apps and Connections work is unchanged. This focused regression complements the existing integration; it adds no separate feature or provider setup path. ## Model Used OpenAI, configured model `gpt-6.1-sol` in Codex. The exact served backend revision and context window are not exposed. Assistance used reasoning, source inspection, Git, native TypeScript checks, and controlled integration tests. A coordinating GPT-6-family agent reviewed boundaries and evidence. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing> Co-authored-by: Codex <noreply@openai.com>	2026-11-06 17:41:04 -06:00
DottaandPaperclip	bb621880e6	fix(storage): preserve agent and skill files during online conversion (#19656 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Agent files and installed skills need durable storage before a legacy volume can be removed. > - Their maintenance commands currently require every writer to stop. > - Normal file edits already use agent row locks and skill publication locks. > - Shared agent heads also remain durable while their agents run. > - This pull request uses those owners to copy settled files online and keeps unresolved recovery states intact. > - The benefit is useful migration progress while unrelated work continues. ## Linked Issues or Issue Description Related: #19642. What happened? Agent-file adoption and skill installation conversion reject apply requests without a global stopped-writer acknowledgement. Agent-file maintenance also reports an active agent as blocked when its shared head is already durable. Live enrollment omits the existing unresolved-local-instruction-copy guard. Expected behavior Copy settled agent files and skill installations under their existing writer locks. Keep old skill directories, versions, and inline bytes. Keep unresolved local copies and unknown historical pin obligations visible to their owners. Steps to reproduce Run agent-file maintenance with `--apply` on the original host with durable storage enabled. A managed agent without active execution is rejected without `--writers-stopped`. The same gate rejects skill installation conversion. An agent that already has a shared head is also reported as blocked while it runs. ## What Changed - Adopt settled agent files under the existing agent row lock. Recognize an existing shared head before checking active execution, without rewriting its projection. - Reuse the local working-copy preservation predicate for first adoption and live enrollment. Keep its original public export. - Prepare skill uploads outside the shared run-pinning lock. Verify the exact local tree and source identity under the cache, slug, and row publication locks. Publish catalog origin history in that same transaction. - Retain old skill directories, versions, and inline bytes. Keep pin retention and volume-removal guards unchanged. - Add PostgreSQL and filesystem coverage for active durable heads, unresolved copies, concurrent source edits, and linked files. Update the maintenance docs. ## Verification - Agent-file durability and live enrollment suites: 16 tests passed with real PostgreSQL and filesystem fixtures. - Managed skill suite: 22 tests passed. Seven final controls also passed: upload-time file edits, source relocation, unknown historical pins, hard-linked files, pinning during a blocked upload, and both rejected/successful catalog-origin publication. - Server source typecheck and focused changed-test typecheck passed. - Module boundaries, diff whitespace, and added-line credential/internal-reference scans passed. - Baseline production rejected both online paths at their stopped-writer guards. Review regressions also reproduced a busy pinning lock during upload and extra catalog history after rejection; their fixes passed real PostgreSQL controls. All expected red results are preserved. - One final-control attempt failed the existing 30-second fixture startup hook before executing tests. A sequential retry passed all four controls without changing guards or timeouts; the failed attempt remains recorded. - Exact head `001a5beed9d0e55c7c178906fbad1f9a47f87788`: [required CI passed](https://github.com/paperclipai/paperclip/actions/runs/42309446426), including typecheck, build, tests, Runner and end-to-end gates. Greptile completed successfully at 5/5 with zero unresolved review threads; GitHub reports mergeable/clean. ## Risks Changed or unsafe sources remain blocked. Storage failures keep the original files and recovery state. This conversion does not settle retained provider copies or historical execution. Unknown pin obligations still block retention and volume removal; this PR alone does not qualify removal of the original volume. ## Model Used OpenAI GPT-6 (Codex). The runtime did not expose a more specific backend model ID or context-window size. Used reasoning, code editing, and tool execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 17:31:04 -06:00
DottaandClaude Opus 5.5	a3f80ea043	fix(onboarding): close the onboarding task when its interview moves to chat (#19643 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - New users start in an onboarding task. The first agent can run the onboarding interview there or in its chat. > - #19512 moves the interview card to the chat when the agent asks it there. The task's card expires. > - The onboarding task then stays `in_review` with no card, run, wake or reviewer. Review attention calls it "stalled". > - Needs you then shows "Choose how this task continues" with Send back to work / Request changes / Mark task done. A new user cannot know what to pick. The card stays after the interview, plan approval and handoff finish in chat. > - This pull request closes the onboarding task when its interview moves to chat, and links the chat in a system comment. > - The benefit is that the user never decides what happens to an internal task. ## Linked Issues or Issue Description Refs #19512 What happened? In a browser walk (Codex API path), the user chose "Interview me" in the onboarding task. Then the user asked in Chat: "Can we do my onboarding interview here in chat instead of in the task?". The chat card worked. The onboarding task stayed `in_review`, and Needs you showed "Choose how this task continues" for it until the end of onboarding. What did you expect? The onboarding task follows the conversation. It closes with a link to the chat. No decision about an internal task goes to the user. Steps to reproduce 1. Start onboarding. Choose "Interview me" on the onboarding task. 2. Open Chat with the agent. Ask to do the interview in chat. 3. Open Inbox. See "Choose how this task continues" for the onboarding task. ## What Changed - `server/src/services/issue-thread-interactions.ts`: after the chat question supersedes the onboarding task's interview cards, the onboarding task is closed (`in_review` → `done`) when no other card is pending on it. A system comment links to the chat. The change writes an `issue.updated` activity entry (`source: "onboarding_moved_to_chat"`). Activity publications and issue post-commit actions run after commit. - The task stays open if it is not `in_review` or if another card is still pending (for example a card for a teammate). - `first-task` skill: tells the agent that the task closes when onboarding moves to chat. The chat then replaces the task for the rest of the workflow (plan, proposal cards, child tasks, results). The agent skips the steps that keep the task `in_review` or make it wait for a child. - `onboarding-assets/first-task/README.md`: documents the behavior. - Tests in `server/src/__tests__/coordinator-handoff.test.ts`: the new red/green case, plus a case that keeps the task open when another card is pending. ## Verification Red on master (fix reverted, same test): ``` FAIL src/__tests__/coordinator-handoff.test.ts > Coordinator canonical handoffs > resolves the onboarding task when its interview moves to the agent chat AssertionError: expected [ 'stalled' ] to deeply equal [] - [] + [ + "stalled", + ] ❯ src/__tests__/coordinator-handoff.test.ts:257:40 256\| // No "Choose how this task continues" stalled-review card for the… 257\| expect(await stalledReviewCards()).toEqual([]); Tests 1 failed \| 3 passed \| 24 skipped (28) ``` The test reads the real Needs you feed (`attentionService.list`). On master it holds a `review` item with `reviewAttentionState: "stalled"` for the onboarding task. That is the "Choose how this task continues" card. Green with the fix: ``` cd server && npx vitest run src/__tests__/coordinator-handoff.test.ts Tests 28 passed (28) ``` Neighbour suites: ``` npx vitest run src/__tests__/onboarding-first-task-admission.test.ts src/__tests__/issue-onboarding-first-task-routes.test.ts \ src/__tests__/issue-thread-interaction-routes.test.ts src/__tests__/issue-thread-interactions-service.test.ts \ src/__tests__/issue-thread-interactions-telemetry.test.ts src/__tests__/coordinator-handoff.test.ts Test Files 6 passed (6) Tests 373 passed (373) ``` - `server` typecheck (`tsc --noEmit`): pass. - `pnpm verify` exits 2 (the README is a broad input). Required CI supplies broad coverage. - Not run: a live browser walk. The Vitest case drives the same service path the chat uses. ## Risks - Low risk. The change runs only when an agent asks a question in its own chat and that question supersedes a pending onboarding-task question (the #19512 path). - The onboarding task becomes `done` without a recorded first-task decision. Later onboarding happens in chat, so nothing waits on this task. A done onboarding parent still accepts child-completion report wakes. - The onboarding task rows are locked before their question rows, the same order question creation on the task uses (Greptile review follow-up). The chat row is locked first; no path locks an onboarding task and then a chat row. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file edits, test runs). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 17:27:39 -06:00
DottaandClaude Opus 5.5	0c904e2fd5	fix(tasks): treat a queued-message Interrupt as a hand-off, not a failed attempt (#19633 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - A user can send a message while an agent works. The message waits in a queue strip above the composer. The strip has an Interrupt button that stops the current turn so the message starts now. > - In a browser walk of onboarding, Interrupt showed a red "Couldn't finish this attempt. Your message is saved." banner with Retry. The thread also showed two rows for the same stopped run ("Stopped for 22s" and "Stopped after 21s"). After about 20 s it recovered by itself. > - A stop that the user asked for is not a failure. The red banner and Retry make a normal action look broken. > - #19369 stopped resource waits from showing as failed attempts. This is the same banner, reached through the Interrupt path. > - This pull request shows the Interrupt hand-off as working, and renders the stopped turn once. > - The benefit is that steering a running agent looks calm and correct: one stopped row, no error, and the queued message starts next. ## Linked Issues or Issue Description Refs #19369 (same banner class, resource-wait path) What happened? While an agent ran, the user queued a message and clicked Interrupt on the queued strip. A red banner "Couldn't finish this attempt. Your message is saved." with Retry appeared at the top of the task. Two rows showed the same stopped run: "Stopped for 22s" and "Stopped after 21s". About 20 s later the banner went away and the message was delivered. Expected behavior A user Interrupt is not a failure. No red banner or Retry. One "Stopped" row, and the queued message starts next. Steps to reproduce 1. Start a task with a Paperclip Runner agent. 2. While it works, send a message so it waits in the queue strip. 3. Wait more than 10 s, then click Interrupt on the queued strip. 4. See the red banner, Retry, and two "Stopped" rows. Paperclip version or commit master `e482f0f632` Agent adapter(s) involved Not adapter-specific (core bug); seen with Claude API through Paperclip Runner. ## What Changed Cause 1 — banner (`server/src/services/task-interruption-presentation.ts`): - The Interrupt route cancels the run with `queuedCommentInterruptQueueId` set to the queued wake, and stamps `queuedCommentInterrupt.requestedAt` on that wake. The projection skips its "intentional Stop" rule whenever a pending wake exists, and treats a pending wake as "working" only for 10 s after the message was queued. A message queued earlier than 10 s before Interrupt, or a cleanup that takes longer than 10 s, fell through to `state: "failed", canRetry: true`. - The projection now reads `queuedCommentInterruptQueueId` from the run and `queuedCommentInterrupt.requestedAt` from the wake (both bounded strings in the existing batched queries). When the cancelled run was stopped for exactly the pending wake, it returns `working` for up to 5 min after the click. Pause, approval, reconnect and other user actions stay first. A pending Retry preparation and exhausted system recovery keep their own presentation. After 5 min, a hand-off that never started shows one Retry, as before. Cause 2 — two rows (`ui/src/components/TaskChatThread.tsx`): - When a native runner turn ends, the live tail stays until the settled turn and its reply comment render, or a 15 s backstop expires. A stopped turn posts no reply, so the tail stayed for 15 s next to the settled turn. That gave "Stopped after 21s" (live tail) and "Stopped for 22s" (settled turn). - A cancelled settling run now hands off as soon as its settled turn renders. Tests: - `server/src/__tests__/task-interruption-query.test.ts`: new embedded-Postgres test drives the real batched query with the exact `resultJson` the Interrupt route writes. It checks `working`/no Retry/no Sentry failure capture, a run stopped for a different queue still fails, and the 5 min bound. - `ui/src/components/TaskChatThread.test.tsx`: new test renders a live native runner turn, then the cancelled run with no reply. It checks there is no live tail and exactly one "Stopped" row. ## Verification Red on master (tests added before the fix): ``` $ npx vitest run src/__tests__/task-interruption-query.test.ts -t "queued-message Interrupt" FAIL task interruption latest run query > shows a queued-message Interrupt as handing off, not as a failed attempt AssertionError: expected { …(8) } to match object { state: 'working', …(2) } - "canRetry": false, + "canRetry": true, - "state": "working", + "state": "failed", Tests 1 failed \| 18 skipped (19) ``` ``` $ npx vitest run src/components/TaskChatThread.test.tsx -t "interrupted native turn" ❯ src/components/TaskChatThread.test.tsx:5528:82 expect(container.querySelector('[data-testid="task-chat-live-transcript"]')).toBeNull(); Tests 1 failed \| 368 skipped (369) ``` Green after the fix: ``` $ npx vitest run src/__tests__/task-interruption-query.test.ts src/services/task-interruption-presentation.test.ts Test Files 2 passed (2) Tests 60 passed (60) $ npx vitest run src/components/TaskChatThread.test.tsx -t "interrupted native turn" Tests 1 passed \| 368 skipped (369) ``` Local `pnpm verify` (affected Vitest selection) result is added in a follow-up comment; required CI supplies typecheck, build and the full suite. ## Risks - Low risk. Both changes are presentation only. Admission, cleanup and Stop gates do not change. - A hand-off that is blocked for a long time now shows no banner for up to 5 min. The queued strip still shows its own wait text ("Waiting for the previous run to finish recovery…") during that time, so the wait is visible. After 5 min, Retry returns. - The thread change applies only to a settling run with status `cancelled`. Failed and succeeded runs keep the existing reply-comment hand-off. - Not in scope: the task status icon changed to `todo` during the hand-off. That is the issue status, not this projection. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file edit, test execution). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 17:18:04 -06:00
DottaandPaperclip	c9391720a1	fix(heartbeat): send saved follow-ups after task completion (#19600 ) ## Thinking Path > - Paperclip manages agents and their work. > - Task messages remain in a durable queue while an earlier run saves its result. > - That result can mark the task done before the user sends the saved messages. > - Send records current user intent, but its wake guard allowed only active task states. > - The old queue therefore remained pending with an obsolete finalization message. > - This change admits that exact Send receipt and reopens the task through the issue service in the successor transaction. > - The agent can receive the saved input once while the completed result remains unchanged. ## Linked Issues or Issue Description What happened? Send on a saved message can leave the message queued after its native predecessor succeeds and commits. The old wait message remains visible. Allowing admission alone also fails: continuation construction rejects a task that remains done. Expected behavior Explicit Send must deliver the saved input once through normal admission. It must retain the previous run and its result. Steps to reproduce 1. Save a follow-up while a native run finishes its result. 2. Let that run succeed, commit its accepted result, and mark the task done. 3. Send the saved follow-up. Before this change, the receipt stays deferred. Related public work: Refs #19542. That change selects recovered workspace files. This change handles the later saved-message admission. Refs #19602 for the independent workflow fixture repair required by CI. ## What Changed - Allow the exact explicit Send wake guard to accept a done task. Keep automatic recovery and cancelled task guards unchanged. - Reopen the task through the company-scoped issue service after admission gates pass. Commit reopening, queue consumption and the successor together. Publish the audit after commit. - Add a native succeeded/committed predecessor regression with unchanged saved work and one successor after two resume calls. - Verify real heartbeat startup under the clicking operator. Keep automatic input, cancelled tasks and paused agents held. - Check workspace preflight for the prospective reopened task. Use the normal company-scoped transition if preflight blocks a done task, so its completion metadata is cleared. - Check a successful native predecessor for its committed accepted result, current ownership and pending cleanup. Keep approvals and questions effective even without a recovery action. - Document the saved Send contract and update the queue test dependency. ## Verification - Final merge qualification for `675914f0fa9b5eb1d2c9c8fcd448e0669d0c4836`: all 56 applicable status checks pass, with no pending or failed checks. [Final CI](https://github.com/paperclipai/paperclip/actions/runs/42307623559) includes typecheck, build, all applicable test and E2E shards, and aggregate verification. Current-head Greptile is 5/5; all five review threads are resolved. Any paused or queued application review is not claimed as a new completed review. - Final base qualification at `675914f0fa9b5eb1d2c9c8fcd448e0669d0c4836`: official preflight passes (1.5s); all six integrated model-normalization cases pass (19.07s); both affected test typecheck layers pass. The integration changes no production server/shared/Runner/adapter inputs, so the preceding 12-case Send proof, two ordinary status guards, and unchanged production typecheck layers remain applicable. Fresh final-head CI supplies broad coverage. - New review regression: both failed and cancelled native predecessors on a task manually marked Done fail before the repair with `continuation_task_ownership_changed` through real `resumeQueuedCommentInterrupt` admission. At `bfb0eff7941090ac3ac43ce4366ba2e3d46612e9`, all 12 affected Send/source/cleanup/workspace/decision cases pass (246.80s), ordinary Done/cancelled rejection cases pass (2; 100.31s), and all seven server typecheck layers pass. History validation uses the prospective reopened status only for an already-verified explicit native Send. No status is changed during a dry run. - Prior `e64e4bb1` CI passed the affected serialized server suites, including the complete saved-message route suite, typecheck and build. General server shard 4 still failed two unrelated stale default-model expectations (`gpt-5.6-sol` versus the new `gpt-6.1-sol` default). The equivalent repair has already merged in #19610 as `8621bcf53ee2e4bbefb7d193add42dd544ca06e8`; the separately prepared #19645 is closed as superseded. This source fix does not weaken or edit those assertions. - Original RED: the committed-source and completed-task startup cases left the saved receipt deferred. - Review RED: Done plus an unrunnable isolated worktree created one successor instead of zero. A successful committed predecessor with pending lease cleanup also created one instead of zero. The new Done-to-blocked path retained stale completion time. Pending approvals and questions each allowed a successor. All failures were captured before their respective repairs. - Integrated production revision `99018264dd1ef002ff3500378184bcd167d9206b`: 417 affected cases pass across the complete native continuation, responsible-user startup, and workspace preflight suites (386 + 18 + 13). These include the original successful startup case and six cleanup/finalization faults that hold the exact saved receipt until settlement, then deliver it once. - `pnpm --filter @paperclipai/server typecheck` passes all seven layers on that integrated head, including the local Runner/SDK dependency build. Module-boundary, whitespace, and added-code-comment checks pass. - Normal master integrations preserve branch history. No force push or assertion weakening was used. Fresh CI and automated review are running on `675914f0fa9b5eb1d2c9c8fcd448e0669d0c4836`, which normally integrates the already-merged #19610 fixture repair after the reproduced inherited CI failure. - The earlier CI attempt failed in five unchanged workflow fixture cases because IMAGE_ONLY was unset. The merged #19602 repairs that fixture. A later inherited duplicate default in that independent fixture has a separate repair; this PR does not edit it. - Follow-up route-fixture RED: three existing native queue delivery cases failed because their fixture marked a run successful without an accepted result or committed finalization. The same three cases pass after the fixture records the complete native success contract (42.08s); delivery assertions and production gates are unchanged. The affected test-tree TypeScript layer also passes. The full route run passed all 119 assertions but failed its 30-second database teardown hook, so it is not recorded as a suite pass. That result is retained. One unchanged bounded rerun was stopped after more than 15 minutes without a final verdict; it is inconclusive, not a pass. - Final integrated head `e64e4bb1`: all 13 focused route and completed-task admission cases pass across two real-PostgreSQL suites (188.95s), including the three repaired route fixtures and every new cleanup/finalization/decision hold. Fresh preflight passes; all seven server typecheck layers pass. Required final CI supplies the broader route and server coverage. - These tests use real PostgreSQL and normal admission, cleanup, queue consumption, and startup services. The startup suite controls the provider boundary. Hosted provider acceptance remains pending normal deployment. ## Risks - Explicit Send can reopen a done task. This is limited to its validated saved receipt and current assignee. A cancelled task remains held. - Reopening uses the existing company-scoped issue transition and shares the successor transaction. A failed admission does not reopen the task. Provider Stop, workspace, approval, pause and budget checks retain their existing owners. - Cleanup and pending decisions retain the done task and saved receipt. An unrunnable workspace uses the existing blocked-task diagnostic. No schema or runtime package changes. ## Model Used - OpenAI Codex, GPT-6 family; exact serving snapshot/context window not exposed; repository analysis, code editing, tool execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 17:15:07 -06:00
DottaandPaperclip	2e340d6235	fix(ui): show unread message counts in agent chat pickers (#19627 ) ## Thinking Path > - Paperclip helps people manage AI agents and their work. > - People switch between agent chats to read replies. > - The agent picker does not show how many messages are unread. > - People need that count before they select a chat. > - The chat list already has the viewer's read state. > - This pull request adds numeric badges from that shared list. > - People can find unread replies without extra requests for each agent. ## Linked Issues or Issue Description What existing behavior does this improve? The chat header agent picker and the secondary sidebar's Add chat picker. Subsystem affected Agent Chat UI, shared task types, and the chat read-state service. Current behavior The pickers do not show the unread message count for each agent. Proposed behavior Show the unread count beside each agent. For example, show 3 for three unread agent messages. Hide the badge when the count is zero. Keep switching available while chat data loads. Reason and benefit People can see which chats need attention before they switch. The count uses the existing shared request and cache. Breaking changes The chat response gains an optional `unreadMessageCount` field. Existing fields and read receipts keep their behavior. There is no database migration. Additional context Related: #19425 changes chat structure. This change adds counts to the current agent pickers. The duplicate search found no PR for these count badges. ## What Changed - Count unread agent comments in the existing grouped read-state query. Keep company and viewer scope. - Exclude human messages, deleted comments, system notices, and fast-response receipts. - Pass memoized count maps to both agent pickers. Use a direct lookup for each row. - Show numeric badges and accessible singular or plural labels. - Test count changes, account isolation, cached opens, search, and large agent lists. - Update the chat stories, fixture data, and development guide. - Make the Slack owner retry-deadline test deterministic after reproducing its CI failure. ## Verification - On current head `05f5fa6201`, the [main CI workflow](https://github.com/paperclipai/paperclip/actions/runs/42305845930) passed, including build, typecheck, all applicable test shards, browser shards, and final verification. The prior server shard 2, 6, and 7 failures do not recur. - [Greptile's current-head review](https://github.com/paperclipai/paperclip/pull/19627#issuecomment-6696161631) is 5/5, its check succeeded, and both previous review threads are resolved. Current head check inventory is complete: 55 successful checks, three intentional skips, and no pending or failed checks. The Paperclip application review summary reports paused; no new completed Paperclip review is claimed. Its existing findings are resolved, and the independent current-head Greptile review is 5/5. - Current head `05f5fa6201` merges master `8621bcf53e` to resolve an import conflict. Master now includes the same model-default test repair. This branch keeps its upstream import, so that file no longer differs from master. - After this merge: all 21 server tests in the Slack owner and model-default suites pass, all 22 tests in the three picker/sidebar suites pass, and token gates pass. The standard preflight still cannot build Rust because `cargo` is absent. Fresh CI must supply the broad checks. - Head `7d125fb8c7` replaces the Slack owner test's real timer wait with the worker's existing clock input. It checks that the intent stays pending one millisecond before the retry deadline and clears at the deadline. All 15 Slack owner integration tests pass locally. Runtime retry behavior is unchanged. - On `4cfb1e9d55`, both review checks gave 5/5. CI [server shard 2](https://github.com/paperclipai/paperclip/actions/runs/42302458830/job/128183671349) failed the Slack credential-lease retry test. The same assertion failed in an unchanged local run. A diagnostic run with added logging passed; the exact timer deviation was not captured. The new test removes the real-time wait assumption and checks the clock boundary directly. The earlier model-default and cleanup failures did not recur in that CI run. - On `a3378dfc40`, both Greptile and Paperclip Review gave 5/5 with no open findings. Typecheck, build, all eight browser shards, and most test shards passed. - That CI attempt failed two server shards. The model-default test still expected `gpt-5.6-sol`, while the current default is `gpt-6.1-sol`. I reproduced both failures locally. Commit `4cfb1e9d55` uses the exported default-model constant; all six tests in that suite now pass. Production model selection does not change. - The retained-cleanup backlog test exceeded its 15-second limit in [server shard 6](https://github.com/paperclipai/paperclip/actions/runs/42299858777/job/128175193320). The unchanged test passes locally with the same limit. Its CI timeout cause is not established. No assertion or timeout was changed. The fresh CI run must recheck it. - Review repair: commit `0db631b215` keeps the existing-chat and unread-agent arrays stable when chat data and the active agent do not change. A new 500-agent test fails before the repair and passes after it. The test also checks count refreshes. - The first CI typecheck failed on duplicate `IMAGE_ONLY` and `TESTOPS_REQUESTED` keys in `tests/runner-e2e/workflow-target.test.ts`. This was present in upstream code. Master commit `e482f0f632` removes the duplicate keys. This branch now includes that commit. All nine workflow-target tests pass locally. - After the latest master merge: all 22 tests in the three affected picker/sidebar suites pass. Before that merge, all 38 focused UI tests passed. - The current head is `05f5fa620134ad199a3d85b66fb2eb93d3bb6077`. All current status checks are complete, the aggregate verify gate succeeds, mergeability is CLEAN, and no unresolved review threads remain. The maintainer has instructed normal merge after checks pass. - After merging current master: 37 focused UI tests pass across AgentChatPicker, CoordinatorAgentSwitcher, AgentConversationsSidebar, useAgentChatRead, and SidebarAgentChats. - After merging current master: all 38 proactive-chat integration tests pass with a real disposable PostgreSQL database. They check counts of 2, 1, and 0 at exact read cutoffs. - The prior revision passed UI typecheck and build, shared build, isolated low-memory server typecheck, token gates, and 30 map/catalog tests. - The 500-agent test bounds index work and count lookups. Search does not rebuild the existing-chat index. - Browser checks used production components with simulated API data. Dark, light, mobile, and Add chat cases passed. Counts of 3 and 12 stayed visible through search. Read agents had no badge. Cached opens made no chat API requests. - Twenty local cached opens took 34.5–57.6 ms, with a 38.4 ms median. This is an unthrottled fixture measurement. - Earlier full local checks could not complete because Rust cargo is absent and the workspace has limited memory. The standard worktree preflight also stops at missing cargo. Required CI must supply broad checks for this PR head. - The first server typecheck ran out of memory while Storybook was active. Its isolated low-memory retry passed. No visual baseline comparison or live provider delivery is claimed. ## Risks - Low risk. The count replaces a boolean aggregate in the same grouped query. Very long chat histories still need that existing scan. - A count can be stale until the existing chat-list cache refreshes. Read receipts still control the unread cutoff. - The change adds no polling, per-agent history request, database migration, or permission change. ## Model Used OpenAI GPT-6 through Codex, with code editing, terminal tools, and browser verification. The runtime did not expose a more specific served model ID or context-window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 17:10:56 -06:00
Nicky LeachandClaude Opus 5.5	913102870d	fix(environments): fill the managed instance default once, under a lock (#19632 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - On a managed instance under `enableManagedSandboxOnly`, boot points the instance default at the managed sandbox when nothing deliberate is set > - That write read the default and wrote it later without a lock. Two overlapping boots, or a boot and an admin write, could overwrite a value the other had just committed > - A revert path also reset that stamped default whenever the mode was turned off, so a platform-filled default could change again later > - A planned plugin-environment default should fill only an unset default and never change a set one, which needs every instance's default to stay set > - This pull request makes the fill one locked transaction and removes the revert, so a filled default is an ordinary set value > - The benefit is a single, race-free rule: boot fills an unset default once, and nothing changes a set default afterwards ## Linked Issues or Issue Description Refs #19551 (the plugin-environment default that builds on this rule) Subsystem affected Server environments: the managed environment bootstrap and the instance settings route. Problem or motivation `applyManagedEnvironments` read the instance default, decided, and wrote it in separate steps with no lock, so a concurrent boot or admin write could be overwritten. When `enableManagedSandboxOnly` was not declared, a revert reset a default the platform had filled. That needed a `managedDefaultStamped` marker, plus a marker clear on every admin default write. A later boot could therefore change a default that had already been set. Proposed solution Fill the default in one transaction that locks the instance default row and rechecks it. Fill only an unset, local or deleted default. Remove the revert and its marker, so a filled default behaves like any other set value. Alternatives considered Keeping the revert and adding locks to it. A revert on a mode change would still move an existing instance's default later, which a fill-only plugin default cannot tolerate. Roadmap alignment This narrows existing managed-environment behaviour. It adds no product surface. ## What Changed - New `server/src/services/managed-instance-default.ts`: `fillUnsetInstanceDefault(db, environmentId)`. In one transaction it creates the settings row if needed, locks the instance default row (`FOR UPDATE`), and rechecks the current value. It writes only when the default is unset, names the local environment, or names a deleted environment. Any other set default is kept. - `server/src/services/managed-environments.ts`: - Under `enableManagedSandboxOnly`, the managed sandbox fill now uses that helper, through a new `fillInstanceDefault` test seam. - The mode-off revert and the `managedDefaultStamped` marker are removed. - The unused `instanceSettings` seam option is removed, along with the `getById`, `update` and `findManagedSandboxEnvironment` methods of the environments seam. - `server/src/routes/instance-settings.ts`: an explicit default write is a plain `update` again. The marker clear and its transaction are gone. - `doc/DEVELOPING.md`: the "Managed environments" section states the fill rule. Existing rows may still carry `managedDefaultStamped` in their metadata. Nothing reads it any more, and it is not part of the managed stock hash. ## Verification - `cd server && npx vitest run src/__tests__/managed-instance-default.test.ts` (embedded Postgres): - The fill writes an unset default, including on a fresh instance with no settings row. It also writes a default that names the local environment, or one whose environment was deleted. - It never changes a default set to an ssh, plugin or other sandbox environment, and leaves a default that already names the target alone. - Controlled overlap: an uncommitted admin default write makes the fill wait, and the fill then keeps the admin's value. This test fails without the row lock. - Two overlapping fills: exactly one writes. - Real boot pass: the default is filled under managed-sandbox-only and stays the same after a boot with the mode off. A set default stays on every boot. - `npx vitest run src/services/managed-environments.test.ts src/services/managed-environments-plugin-entries.test.ts`: the fill runs only under managed-sandbox-only, with the managed sandbox row. It never runs for a plugin environment. Nothing is written when the mode is off. A failed fill does not fail the ensure. - `npx vitest run src/__tests__/instance-settings-routes.test.ts`: an explicit default write is a plain update that touches no environment row. - `npx vitest run src/__tests__/managed-plugin-environments.test.ts src/__tests__/environment-routes.test.ts src/__tests__/server-startup-feedback-export.test.ts src/__tests__/instance-settings-service.test.ts`: pass. - `pnpm --filter @paperclipai/server typecheck` passes. - I did not run the full `pnpm test:run`, `pnpm -r typecheck` or `pnpm build`. CI covers them. ## Risks - Behaviour change: when `enableManagedSandboxOnly` is turned off for an instance, a default the platform filled with the managed sandbox now stays. Local execution is still allowed then, but it is no longer the default unless an admin picks it. - No schema change. Stale `managedDefaultStamped` keys stay in some row metadata and are ignored. - The fill holds the instance default row lock for one short transaction at boot. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, through Claude Code with tool use (file edits, shell, test runs). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 15:01:58 -08:00
Devin FoleyandPaperclip	8621bcf53e	Cancel subscription sign-in requests when leaving agent setup (#19610 ) Back during pending subscription sign-in could accept a late response and reopen agent setup. Abort pending reads, release abandoned sessions, and ignore stale responses. Wait for cleanup before starting a replacement, and preserve sessions belonging to another account or environment. Cover deferred resume/start/prompt responses for both providers, local login cancellation, quick reselection, and browser request abort. Stabilize Runner fixtures and normalization assertions exposed by integration CI. Verified: 483 focused UI tests, six onboarding browser cases, UI typecheck and token gates, Runner fixture checks, and the complete CI run on `221f2e3eb7`. Greptile scored 5/5; all review threads are resolved. Earlier failures and recovery evidence remain in the PR description. Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-11-06 14:31:11 -08:00
DottaandPaperclip	44c60c7221	Verify default skill sync through two real cold sleep cycles (#19191 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Automatic skill sources check upstream every five minutes. > - The domain deadline contract can hand future work to a durable wake alarm. > - A saved deadline or a worker unit test does not prove that a stopped tenant resumes the real producer. > - This pull request checks the default cadence through the actual app and a matching service checkout. > - It checks two cold starts and three distinct persisted skill revisions with real clocks. > - The benefit is repeatable local evidence without changing production timing or authority. ## Linked Issues or Issue Description Refs: #19001, #17008, #17923. This is a test and documentation follow-up to #19001. #19001 is now merged; this PR targets master. Feature code remains in that merged foundation. #17008 covers the earlier source worker change. #17923 covers separate skill-cache retention; this PR does not change that behavior. A scoped search of public PRs and issues found no duplicate default-cadence cold-cycle acceptance test. No matching public issue was found. The enhancement fields follow. Pre-submission checklist This improves checks for an existing feature. It adds no product endpoint or execution authority. The public related-work search is complete. What existing behavior does this improve? Verification of automatic skill-source updates across a durable idle handoff and cold restart. Subsystem affected `server/` test fixtures and development documentation. Current behavior Domain tests cover claims, access and future deadlines. They do not prove the full default-300-second path from ordinary source attachment through signed tenant reports, durable alarms, stopped compute and a changed persisted skill. Proposed behavior An opt-in launcher checks bootstrap first, then two full real-clock cycles. Ordinary CI can run the transport controls without the host fixture. The launcher-specific worker assertion runs only when prewire is requested, and fails closed in that lane. Reason and benefit The test can detect an early refresh, a lost deadline, a stale proof, an ACK after physical stop, a failed cold restart or a producer that does not publish changed contents. Breaking changes None to product APIs, schema, cadence, authorization, claims or retry policy. Existing maintenance-helper defaults stay unchanged. Additional context The opt-in host flow needs an authorized matching service checkout, its normal compiled acceptance bridge, Node 22 and two owned PostgreSQL test databases. The standalone controls do not need that service or PostgreSQL. Roadmap review places this under existing Skills Manager and Cloud deployment verification, not a new core feature. ## What Changed - Add an opt-in default-cadence acceptance test and checked-in launcher. - Reuse real server startup, managed tenant authority and the ordinary project attachment route. Control only the remote Git snapshot boundary; keep actual authorization, scanning, claims and publication. - Use actual signed tenant routes and the matching service's gate, sleep controller and PostgreSQL wake-alarm store. Stop and restart real app children through a controlled local provider. - Preload the Node transport guard before coordinator, worker and child module work. Check worker-local and child PID, thread and source-hash receipts. Keep caught unexpected transport attempts fatal. - Require both persisted refresh times and actual upstream handler timestamps to be at or after each original deadline. - Join physical callbacks during bridge close, stop a late wake-created child, and preserve the original cleanup budget. Add a real IPC/HTTP-child counterfactual. - Accept valid owned HTTP options paths without allowing Unix sockets or unknown origins. Remove the unused App helper export. - Add finite transport and launcher controls. Defer only the launcher-specific parent assertion in vanilla CI. Keep all subprocess controls default-runnable. - Add optional child entry, preload and bounded pool settings to the existing test helper. Preserve its defaults and the real observer test body. - Remove added explanatory comments under the contribution policy. Preserve executable syntax, directives and literals. - Document pinned setup, fresh owned databases, bootstrap and full two-cycle commands in `doc/DEVELOPING.md`. The diff against the intended base contains nine test/helper/launcher files and this one document. It adds no production feature code or Cloud source to the App repository. ## Verification The latest two-file CI repair preserves both original failed jobs. A newer master workflow introduced two required step inputs absent from its extracted-step test environment; the faithful false/false inputs now pass all five existing target-authorization cases, after an actual missing-input counterfactual failed. A controlled socket case also reproduces legal reuse of the released runner port; the fixture now holds that reservation until the preview binds, releases it in finally, and asserts distinct ports. All original retry/result/accounting assertions and timeouts stay unchanged. The original Linux listener owner remains unknown; the control proves this fixture collision mechanism, not that original attribution. Fresh current-head full CI and complete reviews are pending. Current merge-integration candidate: `f6f686103b7bfd0a4ab83085634c622af3896ecb`, a normal merge after #19001 landed, including actual master `1c9767d323cf872a266962b9e6b4bd5cbcb57f1c`. All nine acceptance code files are byte-identical to148a82e5; both parents' documentation additions survive. Every other path equals actual master. The only conflict was the old foundation's startup block; current master was retained in full. Fresh exact-head CI and complete reviews are pending. The local measurements below belong to their explicitly named revisions; no fresh full614-second run is claimed. Current candidate: `148a82e54c502759785407357001c6b50c1439dd`, normally carrying the current base `8c912e1ba4fc66397a7341ade59373af8075068f`. All nine acceptance code blobs remain identical to qualified repair `e092746f4da181c8414c63be71474619a84b97e7`; documentation retains both parents. The matching built service for the fresh bootstrap is `389878364b42dd129835d263f91882c5774d5123`, using Node 22.23.4. The App uses Node 24.22.2 and pnpm 9.15.4. Use the exact controlled setup and commands in [doc/DEVELOPING.md](doc/DEVELOPING.md#controlled-default-cadence-sleep-acceptance). Fresh owned databases and the matching compiled service are prerequisites; the launcher does not create a cluster or build dependencies. - Current affected semantic check passed, using actual native TypeScript 7 and the current generated CI options/references (0.640 seconds). Normal declaration references passed in 11.726 seconds, with source-matched SDK/shared/kernel/Runner prerequisites and the frozen current lockfile. - Current guarded controls: 44 passed, plus the same single explicit real-PG observer deferral, in 4.790 seconds. The executing worker checked its own global PID/thread/preload source hash and matching receipt. Guard hash: `7daf9ae7d2f26e8fcd389b1d560fac1d4e218a8304a2c58e49f646af108165bc`. No unexpected attempts or PG sockets occurred. Vanilla controls on the identical repair files passed 22 cases with one explicit launcher-parent assertion deferral (2.044 seconds). - Real cleanup counterfactual: unchanged old helper failed the actual endpoint assertion in 1.568 seconds: its late HTTP child remained reachable after close returned. The same fixed-helper case passed in 1.485 seconds: close waited, the child exited, the port was unreachable and no IPC error occurred. The isolated driver records only the known legacy closed-IPC error long enough to inspect cleanup; current helper requires zero such errors. All recorded processes and ports were independently checked empty. - Fresh current bootstrap passed once, 16.159 seconds wall / 14.12 seconds Vitest / 12.237 seconds body. The ordinary enabled producer persisted useful revision 1, cleared its lease and retained exactly 300,000 ms between success and original next deadline. A signed current-generation report produced the matching generation/operation-lock SQL ACK before physical child stop (ACK 06:54:02.020 UTC; stop signal 06:54:02.025 UTC). The alarm retained the original D and D-120 eligibility. Both fixture databases ended with zero sessions; the coordinator independently audited and normally removed them. This bootstrap does not execute a due wake or the new v2/v3 timing assertions. - Historical full two-cycle result remains 1/1 passed in 614.213 seconds wall, at measured App `74f7c012ca17ad0d7302d76403f5033affb1430b` and service `2cb54f904031dcf69a32902cd853b43f74d6f301`. Two new processes published distinct immutable v2/v3 at original D+115 ms and D+98 ms. Exact signed SQL ACKs preceded both physical stops; real SQL D-120 claims caused both cold starts. Features, claims and original business deadlines stayed unchanged. - Both historical 10.002-second quiet windows kept app SQL, upstream and outgoing HTTP counts unchanged. Two awake PG sockets remained. Quiet counts do not mean all connections closed. - The stronger timing checks are additional assertions, not retroactively executed results. Offline inspection of the original receipts confirms persisted successes at D+115/+98 ms and earliest upstream handler entries at D+44/+30 ms; all five requests in each refresh group started after D. The exact new pre-sleep request counter and dispatch-before-handler time were not independently recorded historically. - Source review found 19 critical automatic Git producer/runtime files identical to the measured App. The current active cache helper change does not run on automatic Git publication and adds no pool. The matching service carries unchanged measured helper/sleep/controller/alarm seams through its normal master merge. The current bootstrap and focused controls qualify the changed cleanup/guard and parent integration. No fresh current-head 614-second pass or whole-tree identity is claimed. - Runtime is serialized: worker 1, per-pool limits at most 4, configured App potential 10 + shared service pool 2 + one observer against server max 40. Current bootstrap point samples observed at most 3 owned clients plus the observer; those samples are not a continuous aggregate peak. App/service guard peaks were 2 each, with zero remaining sockets at cleanup. Failures remain accounted for. The first historical bootstrap cleanup masked an unknown primary error. Later attempts exposed missing fixture-owned domain metadata and lost origin snapshots; test-only ownership/cleanup corrections preserved the original assertions and timeouts. The new causal driver first lacked a final observation, then encountered sandbox bind EPERM, then exposed the old closed-IPC error; only the final real endpoint failure above is labeled the cleanup counterfactual. An offline dependency miss and a generator-cleared ignored compiler config were retained and corrected without weakening compiler options. The former inherited policy failure (397 findings) is resolved by the genuine current-master base/scanner, with zero findings and seven scanner tests; no unrelated UI site was edited and the gate was not relaxed. Historical real TCP observer coverage passed 1/1 in 3.688 seconds. Comment-only and opt-in-selection carries remain documented in prior evidence; self-hash changes are metadata exceptions with fresh executing-worker checks. The full workspace typecheck, test, build and browser commands were not run locally for this test-only slice. The current base passes its full exact-head CI and Greptile 5/5; this leaf now has all 56 exact-head check runs terminal (54 success, one neutral, one skip), the additional legacy security status passing, complete current Greptile 5/5 and Ponytail 5/5 reviews, and all four review threads resolved across all pages. No live provider or deployed host/database suspension was exercised. ## Risks - The host test requires a matching service checkout and compiled bridge. An unrelated service revision is not a valid substitute. - Controlled localhost Git, S3 and physical process operations do not qualify deployed Railway or Neon suspension, external permissions or provider uncertainty. - The Node guard does not prove native subprocess transport isolation. Only finite local Git metadata and Node version commands were needed. - Two short SQL/HTTP quiet windows do not establish strict 90-second host acceptance. Separate plugin-host LISTEN, MCP wake and other domain gaps remain open. - Explicit test URLs are destructive fixture inputs. Use fresh, owned `_test` databases. The launcher does not start a cluster or choose a database implicitly. - Passing CI and review do not qualify deployed suspension; the current proof boundaries above still apply. > For core feature work, check `ROADMAP.md` first and discuss it in `#dev` before opening the PR. This PR adds verification to the existing feature in its base. ## Model Used OpenAI `gpt-6.1-sol` assisted with implementation, repository review and local code execution. OpenAI Codex, based on GPT-6, performed managing review and integration. The context-window sizes and exact served backend snapshots were not exposed. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change and contains no internal Paperclip ticket id - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green — fresh merge-head gates pending - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups — fresh merge-head review pending - [x] I will address all Greptile and reviewer comments before requesting merge ### Current master reconciliation (2026-11-06) Merged actual master `e28158b353fd274f0322be76d62cd223e0df599f` normally. All nine qualified acceptance code files and the held-port repair remain byte-identical. Upstream independently landed the workflow input fix in #19602; the fixture now uses that exact upstream version once, avoiding duplicate object keys in a squash merge. All other master changes are preserved. The earlier two-cycle local evidence remains source-specific; no fresh 614-second run or deployed suspension is claimed. This new head requires fresh CI and bot reviews. ### Actual combined-tree typecheck failure and repair CI job `128156887548` failed with TS1117 at workflow-target.test.ts:22: two independently landed master fixes declared the same `IMAGE_ONLY` and `TESTOPS_REQUESTED` keys. The complete original log is retained. A further normal master merge now removes the duplicate declarations, retaining exactly one false/false pair and every original assertion. All nine qualified acceptance code paths and the held-port repair remain byte-identical; unrelated master changes are preserved. This is a concrete compiler failure and source repair, with fresh exact-head CI/reviews required. --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 15:55:05 -06:00
Devin Foley Claude Opus 5.5 Paperclip	e482f0f632	fix(server): shed only retryable object-journal reads under load (#19176 ) Client polling can exhaust the shared object journal queue and reject log persistence. Bound ordinary object-log and trace reads, keep persistence and internal verification queued, and return retryable 503 responses when reads are busy. Leave feedback exports unsent until every log page is available. Validation: 72 latest-master affected tests and 10 workflow tests pass locally, with the workflow fixture passing strict TypeScript checks. The preceding broader run passed 280 server tests with two opt-in S3 skips. All applicable checks on `d0a937768d` are green (56 success, three skipped); exact-head Apex review is 5/5 with no unresolved findings. Background persistence still has no fixed queue bound during a prolonged storage outage. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Co-Authored-By: Paperclip <noreply@paperclip.ing>	2026-11-06 13:47:25 -08:00
DottaandClaude Opus 5.5	02a2299173	feat(codex): default Codex to gpt-6.1-sol, half the gpt-5.6-sol price (#19623 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Codex agents run through Paperclip Runner or the legacy `codex_local` harness, and the runner E2E suites run paid Codex cells > - The shared Codex default and many selections preferred `gpt-5.6-sol` > - The reviewed OpenAI rate card in `server/src/services/codex-pricing.ts` (`openai-standard-2026-09-30`) prices `gpt-6.1-sol` at $2 input / $0.10 cached / $10 output per 1M tokens, and `gpt-5.6-sol` at $4 / $0.40 / $20 > - So every default run and every paid Codex eval cell cost twice as much as necessary > - This pull request moves every place that prefers or selects `gpt-5.6-sol` to `gpt-6.1-sol`, and keeps historical facts, rate cards and compatibility fallbacks on `gpt-5.6-sol` > - The benefit is half the token price for new default Codex agents and for the Codex E2E and TestOps cells ## Linked Issues or Issue Description What existing behavior does this improve? The default Codex model for new agents and the Codex model that the runner E2E suites and TestOps plans select. Subsystem affected Adapters (`adapter-utils`, `codex-local`), server native runtime fallback, UI AI connection pools, Paperclip Runner devtools and live evals, runner E2E catalog and TestOps mappings. Current behavior New Codex agents without an explicit model store `gpt-5.6-sol`. The runner E2E `runner-codex` profile, the Reliability `queue-steer` plan cell, the Product regression TestOps mapping and the ACPX Codex qualification selection all use `gpt-5.6-sol`. Proposed behavior They use `gpt-6.1-sol`. Existing agents that store a model keep it. `gpt-5.6-sol` stays selectable. Reason and benefit `gpt-6.1-sol` costs half as much per token on the in-repo reviewed rate card. The current OpenAI pricing page (retrieved 2026-11-06) shows the same short-context rates and also lists long-context rates of $4 / $0.20 / $15 for `gpt-6.1-sol` against $8 / $0.80 / $30 for `gpt-5.6-sol`. Breaking changes None for stored configurations. See Risks. ## What Changed Changed defaults and selections (all now `gpt-6.1-sol`): - `PAPERCLIP_RUNNER_DEFAULT_MODELS.codex` in `packages/adapter-utils/src/paperclip-runner-permissions.ts`. This also changes `DEFAULT_CODEX_LOCAL_MODEL`, `resolvePaperclipRunnerModel("codex", …)`, runner config normalization, `codex_local` → Paperclip Runner conversion, the UI new-agent and agent-config defaults, the composer default model and the public MCP effort lookup. - `codex-local` `models`: the 5.6 entry is now the literal `gpt-5.6-sol`, because it used to be `DEFAULT_CODEX_LOCAL_MODEL`. The list order and contents do not change. - AI connection pool: a new OpenAI member profile uses `PAPERCLIP_RUNNER_DEFAULT_MODELS.codex` instead of a hard-coded `gpt-5.6-sol`. - Paperclip Runner devtools ACPX Codex default and preset, the `runner-codex-sol` local provider smoke, and the `acpx-codex-sol` live workflow candidate. - `tests/acpx-qualification-models.ts` Codex selection (runner E2E and runner acceptance ACPX Codex profiles). - Runner E2E: the `runner-codex` profile follows the default, so all native Codex cells move. This includes the `runner-codex` cells among the 52 standard `everyday-workflows` cells and the seven Codex reliability cells. `runner-codex-mini` stays on `gpt-5.4-mini`. The Reliability plan `queue-steer` `steeringProfile` is now `['runner-codex', 'gpt-6.1-sol']`. Its executionId (`everyday-workflows.runner-codex.daytona.queue-steer`), its fullName and its TestOps case identity do not change. No separate profile is necessary. - `product-regression-testops.mjs`: the declared model for every `runner-codex` case is now `gpt-6.1-sol` (it was already that for `configured-repository`). - `tests/runner-e2e/installed-cli.ts`: the credential-free installed startup proof opens a session on the default model. - `tests/runner-e2e/scheduled-prices.json`: add a `gpt-6.1-sol` price (`openai-standard-2026-11-06`, including 30m/12h cache writes and long-context rates from the OpenAI pricing page). The weekly scheduled Codex cells need an exact price for their model, or admission fails closed. - Storybook: the composer model picker fixtures, the connection pool stories and host, and the agent settings preview show `gpt-6.1-sol` as the default. - Docs: `doc/DEVELOPING.md` (the Codex conversion default) and `tests/runner-e2e/EVERYDAY-WORKFLOWS.md` (local matrix, reliability extras, `queue-steer` row). - Tests that assert the default now expect `gpt-6.1-sol`. New assertions: the default has the `0.159.0` CLI floor, the compatibility fallback is floor-free, the `queue-steer` plan cell runs `runner-codex` on `gpt-6.1-sol`, and all native Codex everyday cells (standard and reliability) use `gpt-6.1-sol`. Deliberately kept on `gpt-5.6-sol`: - Rate cards: `server/src/services/codex-pricing.ts` and `packages/paperclip-runner/src/evals/model-pricing.ts`. Old receipts need them. (The runner eval rate card has no `gpt-6.1-sol` entry. An unpriced candidate is reported as unpriced, not as zero. I did not add a rate there because that card has its own version and source.) - The remote Codex CLI compatibility fallback. A new `CODEX_LOCAL_COMPATIBILITY_FALLBACK_MODEL = "gpt-5.6-sol"` replaces the default in `codexModelFallbackCandidates`. `gpt-6.1-sol` needs Codex 0.159.0 or newer, so it cannot be the floor-free last resort. Fallback behavior is unchanged. - `tests/runner-e2e/provider-usage.ts`: the implicit request cache-policy attribution stays qualified only for `gpt-5.6-sol`. I did not verify that policy for `gpt-6.1-sol`. A `gpt-6.1-sol` response that omits the write-TTL split keeps its cache-write cost unknown, which is the existing behavior for unqualified models. - `tests/runner-e2e/scheduled-measurements.json`: historical `gpt-5.6-sol` measurements. They no longer match the current suite definitions, so the forecast reports them as historical only. The forecast tests now build their synthetic cohorts on the current definition hash. - Model catalogs that list `gpt-5.6-sol` as available (`codex-local`, `cursor-local`, `opencode-local`, Speko voice diagnostics, Pi picker fixture). - The `gpt-5.6` → `gpt-5.6-sol` alias for old stored configs. - Dated or evidence docs: `doc/plans/`, `doc/adapter-model-audit-`, `QUALIFICATION-*`, `doc/connections/MEMORY.md`, `doc/connections/TASK-REVIEWS-VERIFICATION.md`, `tests/runner-e2e/CHAT-FEATURES.md`, the Slack eval evidence, and the S49 soak signature text (the soak failure happened on `gpt-5.6-sol`). - Test fixtures that parse, price, migrate or round-trip an existing `gpt-5.6-sol` value (Rust runner tests, driver tests, pricing tests, stored-config tests, stories that show an existing agent). ## Verification - `npx vitest run` on every server, UI, adapter-utils and codex-local test file that mentions `gpt-5.6-sol` or the default, plus all of `packages/adapter-utils` and `packages/adapters/codex-local`: 5,483 passed. Three timeouts under load (`sandbox-managed-runtime`, `acp.quota`) passed when rerun. - `packages/paperclip-runner`: `vitest run` for `src/eval`, `src/evals`, `src/drivers/acpx`, `src/drivers/codex`, `src/cli`, `src/contracts`, `src/backends`: 2,641 passed. - `node --test tests/runner-e2e/reliability-testops.test.mjs tests/runner-e2e/product-regression-testops.test.mjs`: 28 passed. - `node tests/runner-e2e/run-unit.mjs`: all model, catalog, reliability, scheduled and TestOps tests pass. Failures remain only in `copilot-protection-flow`, `denied-target-isolation-flow`, `pi-controls-flow`, `workflow-target` and `unit-launcher`. These five also fail on unmodified master in this local environment (filesystem watch and git environment), and they do not touch models. - `vitest run --config tests/runner-acceptance/vitest.config.ts`: 19 passed. - `pnpm test:e2e:runner -- --list --suite everyday-workflows --group reliability`: the seven `everyday-workflows.runner-codex.daytona.<case>` cells show `gpt-6.1-sol` with `OPENAI_API_KEY,DAYTONA_API_KEY`. `--matrix-json` gives 16 reliability cells with the same credentials as before. `--list --all` shows no `gpt-5.6-sol` cell. - Typecheck: `adapter-utils`, `adapter-codex-local`, `server`, `ui`, `paperclip-runner` (TypeScript), and `pnpm test:e2e:runner:typecheck`. - `pnpm check:token-gates`: clean. - Not run: paid cells, Rust tests (no Rust change), the browser suites, and the full `pnpm -r typecheck` / `pnpm build` (CI runs them). ## Risks - Agents without an explicit model change model on their next save or normalization, and Paperclip Runner rows with a blank model use `gpt-6.1-sol` at execution. Stored models, including `gpt-5.6-sol`, are never rewritten. No migration is necessary. - `gpt-6.1-sol` needs Codex CLI 0.159.0 or newer with ChatGPT sign-in. The runner pins `@openai/codex@0.160.0`. On a remote target with an older image CLI, the existing compatibility check falls back to `gpt-6-sol` and then `gpt-5.6-sol`, and that probe now runs for default-model agents too. A legacy `codex_local` agent created with the new default on a host with an older Codex CLI gets the existing actionable "upgrade Codex" Test result. - Qualification: the native Codex profile uses `adapter_constant` model qualification, so it follows the default. No qualification gate is dropped. The ACPX Codex qualification selection moves to `gpt-6.1-sol`. Its next qualification run uses the new model. The provider-usage cache-policy attribution stays limited to `gpt-5.6-sol`. - Runner E2E history: the Codex cells change their suite definition hashes, so earlier `gpt-5.6-sol` results are a separate cohort and do not count as current. - The `gpt-6.1-sol` scheduled price uses the file's existing `expiresAt` (2026-11-19). It needs the same review as the other entries before then. - No `.github/` change. The Reliability queue-steer cell keeps the `runner-codex` profile id, so the existing workflow credential and setup steps still match. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (Anthropic), model ID `claude-opus-5-5`, running in Claude Code with tool use (shell, file edits, web fetch of the OpenAI pricing page). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge Related: #19604 (commits the Reliability TestOps identities; this PR changes only `steeringProfile` in the same file). 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 15:35:34 -06:00
Dotta Paperclip Codex	2e35730637	feat(chat): receive sealed Slack events through the current SDK owner (#19019 ) ## Thinking Path > - Paperclip manages AI agents and their work. > - Slack connectors turn authenticated messages into company work. > - Direct callbacks need an awake tenant. > - Cloud can retain authenticated asynchronous events until the tenant wakes. > - Replay must preserve the original callback proof and current endpoint authority. > - Configuration changes must retain registration responsibility through failures. > - This PR enrolls verified Slack identities and drains sealed events through the normal SDK owner. ## Linked Issues or Issue Description Subsystem affected Server chat connectors and signed Cloud connector delivery. Problem or motivation Direct Slack Events API delivery depends on an awake tenant. Broker outages, lost responses and configuration changes must preserve delivery and registration responsibility. Enrollment failures must also leave other connector events able to drain. Proposed solution Stage nonsecret enrollment or retirement intent inside the local configuration transaction. Let the existing connection-event owner perform broker requests and retries. Replay sealed events through the current Slack SDK with their original authentication evidence. Alternatives considered An early acknowledgment for every callback changes command and modal semantics. A second delivery ledger duplicates existing ownership. Reusing a new stack identity to retire old registrations crosses an authorization boundary. A companion Cloud broker change is required. Related setup fixes: Refs #18837, Refs #18878. Kickoff searches found no competing async enrollment consumer. The ROADMAP's connector setup and access controls stay in place. ## What Changed - Persist enrollment and credential-free retirement intent before local configuration commits. Broker failure does not make a completed pause or removal look unsuccessful. - Bind company, endpoint, connection, public ID, generation and authorized actor. Preserve lost-response responsibility and recover only the supported same-instance configuration outage. - Permit an explicit move to a new authorized stack only after confirmed old retirement. Keep the old identity until the new broker receipt matches the current intent and generation. - Show permanent signing-credential failures as attention. Preserve transient retries and reject uncertain inbound authentication before SDK admission. - Drain events and acknowledgments before enrollment reconciliation. Persist a bounded retry on the exact busy intent so other due endpoints can enroll or retire. Preserve current custody under concurrent changes. - Stage retirement in the Apps removal transaction too. Archive the endpoint and connection, retain the original subject and instance, then revoke credentials. Retirement does not need the deleted Slack credential. - Acknowledge a malformed sealed Slack receipt as invalid once. Keep actual broker outages on the retry path. - Run the real configuration, SDK ingress, generation fences and public Apps removal paths in the default SQL suite. Keep Vitest transport helpers outside production services. - Preserve original callback bytes, signature, time and public URL evidence. Use the existing transactional chat-delivery ledger for redelivery. - Provide a checked-in prewire guard and launcher using the current Node executable. Add broker, worker, SDK, SQL and controlled-process acceptance. - Stabilize AI account summaries by ID and grant ID. This fixes a CI migration-repeat equality failure without changing SQL, authorization or reconnect behavior. - Open the four configured database connections before the enrollment test freezes its clock. Keep all 35 original assertions and all timeouts. - Measure status-card dispatch entry before its mock does database work. Keep the original strict 500 ms bound and check any still-armed earlier deadline. Bound that fixture's pool to four connections. ## Verification - Candidate: `54fa701d41b4ced62f4ff338d39b946015670ae3`. Normal merges preserve master `47027e7` and all owned changes. A read-only merge assessment against fetched master `cb5d7e44` is conflict-free. The only overlapping file has separate AI test additions. The branch has not merged that later master; fresh remote checks must validate the current PR test tree. - At this head, normal server production typechecking and declaration emission passed in 35.24s. The affected actual CI test shard passed in 6.27s. Other unchanged CI layers, 92 affected pure controls and 12 guard controls retain their source-matched results. - All 15 default real-PostgreSQL cases passed in 34.66s, with no skips or filtering. They cover normal configure and SDK ingress, rollback, stale generations, malformed receipts, busy-lease fairness, concurrent intent custody, unknown errors, and public Apps removal. Removal tests check actual credential revocation, retained retirement through outage and repeat removal, one ordinary retry and confirmed retirement, transaction rollback and foreign-company refusal. - The public removal counterfactual failed before the fix: archival and credential revocation passed, but the inactive intent was absent. The same test source passed after the staging call. Separate counterfactuals also reproduced busy-endpoint starvation and malformed-receipt retries. Original failures remain retained. - The paired broker/SDK/process journey passed in 31.98s at ancestor `0324c011`. It preserves callback proof, lost-ACK redelivery, exactly one comment, one delivery and one successful controlled process. Its test, producer, SDK, guard and runtime bytes are unchanged. The new public Apps archive branch is not called by that journey; its evidence carries on that checked basis. - The new SQL cases need no companion Cloud checkout. Normal CI runs them with the repository fixture. For an existing dedicated PostgreSQL fixture, set `PAPERCLIP_SLACK_OWNER_TEST_DATABASE_URL` and `PAPERCLIP_TEST_POSTGRES_ADMIN_URL`, then run `node server/scripts/test-slack-cloud-owner.mjs` with Node24. The optional paired launcher still requires the normally built companion Cloud checkout. - Every pool was bounded to four or fewer connections. The final 15-case run sampled three runtime clients plus one observer. The independent final audit found zero clients and all five recorded processes absent. Earlier closing-socket observations are kept separate from final cleanup. - Earlier unrelated CI failures are documented: [enrollment timeout](https://github.com/paperclipai/paperclip/actions/runs/41944373937/job/126973483754) and [status-card observation bound](https://github.com/paperclipai/paperclip/actions/runs/41944373937/job/126973483843). The enrollment test retains all 35 assertions and its original timeouts. A controlled connection probe established the frozen-clock mechanism; the original remote access history remains unknown. The status-card fixture measures dispatch entry with the original strict 500ms bound. A real delayed-dispatch control fails that bound. - Full local workspace typecheck, build and test inventory was not run because its embedded-cluster and wider resource paths exceed this local lease. Exact-head remote CI now passes: 54 successful checks, three skips and Ponytail's neutral check, with aggregate verification green and the legacy security context independently confirmed successful. Both complete bot reviews are 5/5, with all 13 paginated review threads resolved. Seven execution jobs originally reported runner shutdown; one diagnostic rerun recovered them and the dependent aggregate check. Original logs remain retained, and the underlying shutdown cause is unknown. This is recovery evidence, not a product-fix claim. - Provider responses are controlled. This proves local SQL, broker, SDK and process behavior. It does not prove live Slack, agent tools, synchronous callbacks or deployed suspension. ## Risks The companion broker is required. Outage or conflicting authority can leave explicit pending or attention responsibility. A move between stacks requires confirmation from the old authorized owner first. Synchronous commands, actions, modal validation, options and challenges retain their deadlines and endpoint blocker. This slice grants no full Slack endpoint sleep exemption. No deployed suspension or savings claim is made. ## Model Used OpenAI `gpt-6.1-sol` workers through Codex, with reasoning, repository tools and code execution. The served backend revision and context limit are not exposed. A GPT-6-family coordinator performed technical review and integration. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing> Co-authored-by: Codex <noreply@openai.com>	2026-11-06 15:22:51 -06:00
DottaandClaude Opus 5.5	876a2e4857	fix(recovery): recover a prior boot's unstarted run at lease expiry (#19578 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - After a server restart, a task must continue automatically and promptly when that is safe > - A user clicks Retry, and the server is killed before the new run starts a process > - Recovery leaves that run `running` until the orphan reap's five-minute stale-age threshold passes > - The user sees nothing happen for about five minutes > - This pull request recovers that run when its controller lease from the earlier boot expires > - The benefit is that the automatic retry starts about one lease period after the kill, not five minutes later ## Linked Issues or Issue Description What happened? A Retry admitted a legacy run. The server was killed before the run recorded a process. The run kept a controller lease from the killed boot and `executionStage: preparing`. After the restart, live recovery skipped the run until `updatedAt` was five minutes old (`server/src/services/heartbeat/recovery.ts`, `staleThresholdMs ?? 5 * 60_000`). Restart-chaos measured 5 m 01 s from run creation to `process_lost` and the automatic retry, twice (4 m 20 s after the server was ready). Expected behavior Recovery should claim the run as soon as the earlier boot's lease expires. The lease is the only liveness proof for a legacy controller, and no process was ever recorded. Paperclip version or commit `master` at `71313207eb`. Steps to reproduce Click Retry on an interrupted task, then SIGKILL the server within about one second. Observe the retried run stay `running` for about five minutes. ## What Changed - `recoverActiveExecutions` skips the stale-age gate for a legacy `running` run when all of these are true: its controller lease is expired (checked by the existing live-lease gate), the lease belongs to an earlier boot (`controllerBootId !== legacyControllerBootId`), the run is still `preparing` (no adapter dispatch), and the run has no recorded `processPid` or `processGroupId`. - The stage and process fields committed with the revocation are checked again. If the old controller dispatched before its lease expired, the run keeps the age guard. - The existing revocation (`revokeExpiredLegacyControllerWithState`) still requires the expired boot ID, so a concurrent renewal wins. The existing `restartInterruptedBeforeDispatch` rule still decides the single process-loss retry. - `doc/execution-semantics.md` describes the rule in section 10. - Two regressions in `server/src/__tests__/live-recovery-work.test.ts` (embedded Postgres, real residual-recovery workers). ## Verification - Red before the fix: `recovers a fresh run from a prior boot at lease expiry ...` failed with `expected 'running' to be 'failed'`. - Green after the fix: `pnpm exec vitest run src/__tests__/live-recovery-work.test.ts -t "stale-age\|fresh foreign controller\|suppressed live runs"` passes 4/4. The run fails `process_lost` with `executionRecovery.providerWorkStarted: false` within seconds of lease expiry, and exactly one retry runs to `succeeded`. - Race case: a DB trigger makes the row committed with the revocation say `dispatching` while the scan saw `preparing`. The run stays `running`, and no retry is created. - Guard cases: a run with an expired lease of the current boot, a run from an earlier boot with a recorded process, and a run from an earlier boot that was `dispatching` (for example a remote HTTP adapter) all stay `running` after four minutes. - Related suites: `heartbeat-process-recovery`, `heartbeat-observed-workspace-preservation`, `native-startup-cleanup-recovery`, `services/heartbeat/recovery.test.ts`, full `live-recovery-work`. - Live K8 probe (restart-chaos supervisor, local instance, ACPX Claude agent): the first run fails with a provider error, the probe clicks Retry, and the server is SIGKILLed 100 ms later, before the Retry run records a process. The probe then measures how long the Retry run stays `running`. \| Build \| Retry run created to `process_lost` \| Server ready to `process_lost` \| Automatic retry run \| \| --- \| --- \| --- \| --- \| \| Base (`71313207eb`) \| 300.1 s \| 293.5 s \| started at 5 m 00 s \| \| This PR \| 60.1 s \| 54.5 s \| started 0.1 s after `process_lost` \| The claim now happens when the killed boot's 60-second controller lease expires. In this probe the restart took about 6 s, so the claim came about 54 s after ready. In the restart-chaos waves, kill to ready took about 41 s, so the claim would come about 20 s after ready. - A chaos wave with the `retry` trigger (`w-20261106T195715-s2727`, branch #19513 merged locally) missed the trigger: no Retry was offered because recovery completed the task automatically (claude t01 PASS). The probe above makes the Retry-then-kill timing deterministic. ## Risks - Low risk. The change only removes an age wait. Ownership is still proved by the expired lease of another boot. A rolling deploy or hot restart whose earlier instance is still alive keeps renewing its lease, so the run is not touched. - Runs with a recorded process, and dispatched runs, keep the five-minute guard. A process on another host or a remote request is not inferred dead. - Native runs are not changed. Their ownership is in `native_run_finalizations` and startup native recovery. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file edits, test execution). Extended reasoning enabled. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 15:17:58 -06:00
DottaandClaude Opus 5.5	586caa5f49	fix(native): hand a shutdown-interrupted startup to the next controller (#19596 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - The native Runner path runs an agent turn through a runnerd process that the control plane owns > - A restart of the control plane must keep the task going without a click, if that is safe > - A graceful shutdown closes the HTTP listener first and then drains the heartbeat runs > - A native run that was between provider dispatch and runnerd backend setup in that gap failed with `local_runtime_api_listener_unavailable`, so the drain did not see it as running and did not suspend it for the next boot > - This pull request makes that failure a restart detach: the run stays running with its claim, the drain suspends it, and the next controller adopts it > - The benefit is that a SIGTERM in this window no longer spends a failure and a same-run retry, and the task continues by itself ## Linked Issues or Issue Description Refs #19530 (the same restart window, before provider dispatch). What happened? In a restart chaos replay (ACPX Claude, local runner), a restart-recovered run got SIGTERM while `createRunnerdBackendWithinSessionClaim` built the runnerd backend. `resolveLocalRuntimeApiUrl()` threw `local_runtime_api_listener_unavailable` because shutdown had already closed the listener. The run settled `failed`. The drain then logged `restartSuspendedRunIds: []`. The next boot used the one bootstrap retry. A second SIGTERM hit the same window, and the task stopped at a manual Retry. Expected behavior A run that this process cannot continue because it is shutting down is handed to the next controller. It is not settled as failed. Steps to reproduce 1. Start a native Runner task and restart the server (K1) so that startup recovery resumes the run. 2. Send SIGTERM while the resumed run sets up its runnerd backend. 3. See the run fail with `local_runtime_api_listener_unavailable` and the drain report no suspended runs. ## What Changed - `server/src/runtime-api.ts`: `resolveLocalRuntimeApiUrl()` throws the new `LocalRuntimeApiListenerClosedError` when this process had a listener and closed it (the `listening` flag is false or the `close` event fired). The error message stays `local_runtime_api_listener_unavailable`, so the existing classifiers do not change. A listener that never started still throws the plain error. - `server/src/services/native-runtime/native-session-executor.ts`: for a local Runner, the startup failure handler treats `LocalRuntimeApiListenerClosedError` like a restart detach. It stops lease renewal, keeps the claim and throws `NativeControllerDetachedForRestartError`. Heartbeat already keeps such a run `running`. A remote Runner keeps the same-run bootstrap retry from #19462, because startup recovery has no process evidence to adopt a claimed remote startup (`native-runner-restart-preparation.integration.test.ts` still covers that path). - `doc/execution-semantics.md`: describe the new rule. - Tests: a heartbeat regression that closes the real runtime API listener before backend setup, then drains and restarts; a unit test for the closed versus never-started listener; two `vi.mock` factories keep the real module exports. ## Verification - Red before the fix: `pnpm exec vitest run src/__tests__/heartbeat-native-restart-suspension.test.ts -t listener` failed with `restartSuspendedRunIds: []` and the run `failed` (`local_runtime_api_listener_unavailable`, `bootstrap_retry`). - Green after the fix: the same test passes. The drain suspends the run, the run stays `running` with `awaiting_runner_reattach`, and the next controller claims it and reaches the backend once. - `pnpm exec vitest run src/__tests__/heartbeat-native-restart-suspension.test.ts src/http/server.test.ts src/__tests__/server-startup-feedback-export.test.ts src/services/native-runtime/native-session-executor.test.ts src/services/native-runtime/native-recovered-owner-followup.test.ts`: 1333 tests pass. One timing test in `native-session-executor.test.ts` failed once under host load average ~30 and passed alone. - `pnpm exec tsc --noEmit -p server/tsconfig.json`: pass. - Live, restart chaos harness, local ACPX Claude (`claude-sonnet-5`), `--rc06-masked`, this branch at `a35ec8b845` (before the remote-scope commit `5dc63f24fb`, which does not change the local path). The original replay schedule no longer reached the window, because the task now ends before the second fault. So I moved the faults earlier: K1 at 10 s, then K2 as soon as the recovered server is ready. - K1 + K2, 2 runs: both runs PASS with R1 pass. Each drain logged `restartSuspendedRunIds: [<run>]` (before: `[]`). The next boot classified the run `resume_dead_runner` and the same run completed. No `local_runtime_api_listener_unavailable` failure. - K1 + K2 + K2, 2 runs: each drain suspended the run. The fourth boot blocked the run with `native_restart_recovery_blocked`, because three restart claims used the three-attempt recovery budget. Retry then completed the task (R3 pass). That budget count is a separate open item. This pull request does not change it. ## Risks - Low. The new path applies only after this process closed its own listener, which happens only during shutdown. The run then waits for the drain or the next boot, as for the other restart detaches. - If shutdown does not reach the drain (for example SIGKILL after the listener closes), the run keeps its live claim until the claim expires. Then startup recovery handles it as a hard restart. This is the same as any run that is running at a hard kill. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, through Claude Code with tool use (shell, file edits, GitHub CLI). Extended reasoning on. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 15:10:44 -06:00
DottaandClaude Opus 5.5	dc1adb7209	fix(tasks): advertise a native resume only while the scheduler would claim it (#19523 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - When a native run cannot reattach to its provider after a control-plane restart, the native coordinator resumes the same run by itself after about 30 seconds, up to three attempts > - The run projection shows this correctly as "Retry scheduled" with only `inspect_run` permitted > - The task interruption shown on the issue read only system-recovery evidence and `scheduledRetryAt`. It never read the coordinator's `nextAttemptAt`, so it showed `state: failed, canRetry: true` and offered a manual Retry while the automatic resume was still pending > - The two surfaces disagree, and a user click can race the automatic resume. That is one way a Retry gets parked behind "execution checks" > - #19503 made the task interruption read the coordinator's pending resume time. Review found that the coordinator also pushes that time forward while it defers an ineligible candidate, so a reassigned or blocked task could hide Retry for good > - This pull request advertises the resume only while the scheduler's synchronous admission checks pass > - The benefit is one consistent story: while the automatic resume is pending, the task shows it with its time and no competing Retry. If recovery ends, exactly one Retry returns ## Linked Issues or Issue Description Follow-up to #19503. Refs #19487, #19497. What happened? A native run failed with `native_session_interrupted` after a restart. Its finalization row had `phase: retryable_failure` and `nextAttemptAt` 30 seconds later. The run projection said "Retry scheduled". The issue interruption said `state: failed, canRetry: true, nextRetryAt: null`. What did you expect? While the automatic resume is pending, the task shows that it is resuming, with the time, and no Retry. When the resume fails for good or never runs, exactly one Retry that works on the first click. Steps to reproduce 1. Run a native task and restart the control plane so the native attach fails. 2. Read `GET /api/issues/:id` while the coordinator's resume is pending. 3. Observe `interruption.canRetry: true` while `execution.phase` is `retry_scheduled`. ## What Changed - `task-interruption-presentation.ts`: the pending-resume read now also requires the task to be in a resumable status, assigned to the run's agent, not owned by another execution or checkout run, the run's question not cancelled, and the agent invokable (not paused, terminated or pending approval). These mirror `readNativeSessionResumptionAgent`. - `native-session-resumption-statuses.ts`: one shared status set used by both the scheduler admission and the read model. - No time cap: provider cooldowns and remote backoff schedules stay visible as a pending resume. ## Verification - `task-interruption-query.test.ts` (embedded Postgres): a reassigned task and a blocked task show one Retry; a paused agent asks to be resumed; in each case `readNativeSessionResumptionAgent` also refuses the candidate, so the read model and the scheduler agree. A resume an hour out (provider cooldown) stays `retry_scheduled` with no Retry. Red on master for the reassigned case. - All 57 presentation and query tests pass. Server typecheck passes. ## Risks - Low risk. Read model only. - A review-assignment resume (reviewer is not the assignee) shows Retry instead of the pending resume. That is the safe fallback. - Deferrals for a pause hold, dependency readiness or a competing run are not mirrored. Those gates also hold a manual Retry, so no competing action appears. - The `POST /issues/:id/retry` endpoint is unchanged. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file editing, embedded Postgres tests). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 14:38:29 -06:00
DottaandClaude Opus 5.5	8c7e86e73b	fix(workspaces): let non-git workspace directories pass close readiness (#19553 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Execution workspaces hold the files that agents work in, and a person can close a workspace when the work ends > - Before close, Paperclip checks "close readiness" so that it does not destroy unsaved git work > - The check assumed that every existing workspace path is a git repository > - A shared or `local_fs` workspace can be a plain directory. For that directory, `git rev-parse` fails, and the check reported "Paperclip could not verify the workspace git status" as a blocking reason > - So a person could never close these workspaces > - This pull request treats a path with no `.git` marker as a workspace with no git state > - The benefit is that plain-directory workspaces can close, and real git errors still block close ## Linked Issues or Issue Description What happened? `GET /api/execution-workspaces/:id/close-readiness` returned `state: "blocked"` for a `shared_workspace` with provider `local_fs`. The workspace directory existed but was not a git repository. The only blocking reason was "Paperclip could not verify the workspace git status. Retry before destructive cleanup." Steps to reproduce 1. Create a project workspace and an execution workspace (`shared_workspace`, `local_fs`) whose `cwd` is a plain directory that exists. 2. Request close readiness for the execution workspace. 3. See `state: "blocked"` and `isDestructiveCloseAllowed: false`. Expected behavior The workspace is ready to close. It has no git state to lose. ## What Changed - `inspectGitCloseReadiness` in `server/src/services/execution-workspaces.ts`: when `git rev-parse --show-toplevel` fails, the provider is not `git_worktree`, and no `.git` entry exists on the path or on an ancestor, return `git: null` with `statusInspectionSucceeded: true`. - Added `hasGitMarkerInPathOrAncestors`. The check uses the filesystem, not git error text, so it does not depend on the git locale. It uses `lstat`, so a dangling `.git` symlink still counts as a marker. Any error other than `ENOENT` counts as a possible marker, so the check fails closed. - For a plain directory, the result is a git readiness object with `repoRoot: null`, the same shape as the missing-path case. `createdByRuntime` stays on it, so close readiness still plans `remove_local_directory` for a runtime-created directory. - Added two regression tests in `server/src/__tests__/execution-workspaces-service.test.ts`: - A plain directory is `ready_with_warnings` with `git: null` and no blocking reasons. This test failed before the fix (`state: "blocked"`). - A directory with a broken `.git` pointer is still `blocked` with the git-status reason. - A directory with a dangling `.git` symlink is still `blocked`. This test failed before the second commit. - A runtime-created plain directory plans `remove_local_directory`. This test failed before the second commit. - A plain directory nested under an ancestor with an unreadable `.git` marker is still `blocked`. This test covers the ancestor walk. ## Verification - Red/green: the new non-git test failed on `master` (`state: "blocked"`) and passes with the fix. The broken-git test passes on both. - `cd server && npx vitest run src/__tests__/execution-workspaces-service.test.ts src/__tests__/execution-workspaces-routes.test.ts`: 2 files, 101 tests pass. The five close-readiness tests also pass after the third commit. - `pnpm verify` selected 2,223 of 2,991 test files because this service has many importers. I did not run that near-full suite locally. Required CI supplies that coverage. - `tsc -p server/tsconfig.json` shows no errors in the changed files. ## Risks - Low risk. The change only affects paths where git inspection already failed. - With `git: null` removed, the result uses the same `repoRoot: null` shape as a missing path. Existing code already handles that shape. - `git_worktree` workspaces still fail closed. A real repository that git cannot read (corrupt, missing git dir, ownership refusal) still has a `.git` marker, so it still blocks. - The terminal-workspace sweep also uses this inspection. With no repo root there is no HEAD SHA, so delivery cannot be "merged", and the sweep still skips these workspaces. Only an explicit close by a person proceeds. - A `.git` on an ancestor across a filesystem boundary makes git report "not a repository" while the marker check finds it. That case stays blocked, which is the safe result. ## Model Used - Anthropic Claude Opus 5.5 (`claude-opus-5-5`) in Claude Code, with tool use (shell, file edit, test runs). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 14:33:09 -06:00
DottaandClaude Opus 5.5	4938a930d6	fix(budgets): stop pending or restart-lost accounting from hard-stopping companies (#19497 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Budget admission and run accounting decide whether any task in a company may start > - A control-plane restart that kills a provider turn mid-flight can lose that turn's final usage receipt. ACPX Claude reports usage only at turn end, so the receipt never becomes complete > - Accounting kept such a run "awaiting receipt" forever. Every later admission in the company then refused with "completed runs await accounting", and the budget evaluator opened a hard-stop incident that paused the company. No product action could clear it: raising the budget refuses while runs await accounting, accounting retry finds no new data, and allowing unpriced usage does not apply to pending runs > - Restart chaos waves showed this as a dead task Retry, a new task blocked as "over budget" at 22 of 500 cents, and a hard stop that cancelled queued Retries and an unrelated running task > - This pull request closes the debt from evidence: once native recovery reaches terminal failure and the provider stop is verified, the known part settles as an unpriced charge. Until then the run holds new work as recovering accounting, without a hard stop > - The benefit is that a restart no longer bricks a company with a hard budget. The operator's unpriced-usage policy decides admission, and unknown spend still never shows as zero cost ## Linked Issues or Issue Description Refs #19399, #19426, #19431 (earlier restart Retry fixes; this is the budget layer under the same journey). What happened? After a control-plane SIGKILL during an ACPX Claude turn, the killed run ended `native_provider_terminal_failed` with `usage_json.accountingReceiptReady = false` and `cost_accounting_pending = true`. It stayed that way for good. The task Retry was accepted but parked with `budget.blocked` ("Company cannot start work while completed runs await accounting"), which the UI showed as "Retry is waiting for execution checks". A new Codex task was escalated to `blocked` with "the original recovery target is over budget" while the company had spent 22 of 500 cents. After the next restart the budget evaluator opened a hard-stop incident (`amount_observed` 22.08 < 500), paused the company and cancelled both queued Retries. What did you expect? A restart should not stop the company. Unknown usage from a stopped provider should become an honest, operator-governed state, and Retry should start a run. Steps to reproduce 1. Give a company a hard-stop budget. 2. Start an ACPX Claude task and SIGKILL the control plane mid-turn, so the resumed turn fails. 3. Observe the run stay `cost_accounting_pending` with an incomplete receipt, and every new admission refuse. ## What Changed - `budgets.ts`: completed runs that await accounting now hold new work only. They no longer open a hard-stop incident, pause the company, place an agent budget hold, or deliver Stop to running work. Only spend at the limit, or unpriced usage under a blocking policy, does that. Chaos wave A and a pre-restart lifecycle cell showed the wedge: a run that had just succeeded stayed terminal for about 100 ms before its charge committed, the evaluator paused the company in that gap, and the Stop it delivered left a cancelled run whose receipt never closed. - `run-cost-accounting.ts`: a failed native run is a closed receipt source only with positive stop evidence: its finalization is `terminal_failure` with no result (not a finalization-deadline failure), it no longer records a process identity, and its latest server-authored process event is `native.local_process_stopped`. One SQL predicate serves the reconciler scan and settlement. Accounting now settles its known part as an `unpriced` charge instead of waiting forever. The background reconciler selects these runs. The reservation settles. Other incomplete receipts stay pending as before. A cancelled run keeps its existing accounting drain, so Stop recovery can still deliver its final receipt. - `native-local-process-stop.ts` and its transactional callers: releasing a verified local or remote process stop wakes the accounting worker, so the closed debt settles promptly. - `usage-receipts.ts`: a charge settled without its receipt is marked `accountingReceiptUnavailable`. A later complete receipt that carries its own price (including a rate-card price) corrects it. A receipt with no own price and unchanged usage does not reopen the hold. - `budget-exposure.ts`: the blocking unpriced-usage policy no longer counts a charge settled from a lost receipt with no observed usage. That is unknown, not "usage with no reliable price". Complete receipts that report no tokens (for example Cursor Cloud) still count. A live wave under the default policy hard-stopped its company at 0 of 500 cents on one such charge. - `budget-exposure.ts`: a terminal-failure native run with an incomplete, not capture-failed receipt counts as recovering accounting. Admission holds new work ("New work must wait while a native run recovers its unfinished accounting") without a hard-stop incident, a company pause or cancellation of running work. - `recovery/service.ts`: the stranded-work comment names the actual budget gate instead of always saying "over budget". - `task-interruption-presentation.ts`: a Retry parked by `budget.blocked` says it waits for the budget, not for "execution checks". - `doc/SPEC-implementation.md`: documents the closed-receipt contract next to the existing incomplete-receipt rule. ## Verification - New embedded-Postgres suite `server/src/__tests__/closed-native-receipt-accounting.test.ts`, built from the persisted shape of the killed run: - Red before the fix: the run stayed pending, `reconcileRunCosts` accounted 0, and admission refused with "completed runs await accounting". - Debt stays pending with no verified stop event, with a finalization-deadline failure, with a recorded process identity, and while the coordinator can still produce a result. - Under a blocking policy, a lost receipt with no usage does not hard-stop; one with tokens still blocks. A run with a still-recorded process identity paused the company with a hard-stop incident. - Green after: the stopped run settles as one `unpriced` charge with the reservation settled; admission is clear under an `allow` policy and names "no reliable price" under `block`; a run with a recorded process identity or a non-terminal coordinator stays pending; the closing window holds admission without an incident or pause; a final receipt that arrives later still enters late-receipt reconciliation. - Pending accounting from a run that just finished holds admission with no incident, no pause, no enforcement version bump, and an unrelated running run stays running. Red before the `budgets.ts` change: the company was paused. - A later priced receipt corrects the unavailable charge to its price; an unpriced one leaves it settled. - `task-interruption-presentation.test.ts`: the `budget.blocked` Retry wait message. - `budget-enforcement-scoping.test.ts`: the accounting-hold release cases now assert that pending accounting places no budget hold and that admission clears after settlement. - Existing suites pass locally: all budget suites (control publication, enforcement scoping, enforcement worker, evaluation work, live attempts, history, service, live usage, budget cancellation), native-stop-accounting, explicit-native-continuation, native-case-handoff, cost-data-quality, heartbeat-cost-accounting, run-cost-accounting, native-usage-delivery. Server typecheck passes. - Runner e2e cell `lifecycle-baseline.runner-codex.local.lifecycle-dependency-restart-neutral`, which failed on master with a false agent budget hold before any restart, passes on this branch. - Live restart-chaos wave (local runner, K1 SIGKILL, codex + 2× ACPX Claude + OpenCode, 4 restarts, company policy `unpricedUsagePolicy: allow`) with the accounting fix: every killed run settled (no `cost_accounting_pending` left), the company stayed active, the OpenCode Retry completed the task on the first click and the Codex Retry finished its task. That wave also exposed a 1-second false hard stop in the closing window (observed 0 of 500 cents) that cancelled the running OpenCode turn; the `budget-exposure.ts` change fixes it and has its own red/green test. - Confirming live wave with both commits, same plan (2× ACPX Claude, OpenCode, Codex; 4 K1 restarts): both Claude tasks passed R3 (one Retry click started a run that completed the task), OpenCode passed R1 (continued by itself), and the Codex Retry completed its task. Zero budget incidents, zero cancelled runs, zero runs left pending accounting, company active; three killed runs recorded as `unpriced` charges. The only open signature is an unrelated workspace-durability check (`R2:pre_fault_files_lost` on the Codex long-bench task). ## Risks - Behavior change: a stopped native run's lost receipt now becomes an `unpriced` ledger charge with its known usage. If that usage has tokens, the default `block` policy still stops new work with "recorded usage has no reliable price". If it has none, the charge is visible in Costs but does not stop the company. Other unpriced charges keep their existing effect. Unknown spend is never recorded as a priced receipt. Budget enforcement stays best effort, as the spec says; killed OpenCode and Codex requests already lose their in-flight usage. - Remote runs and runs without a verified local stop keep their debt pending (an admission hold only, no hard stop). - Contract change: pending accounting no longer pauses a scope or stops running work. New work still waits with "completed runs await accounting". A scope's real hard stop (spend at the limit, or blocked unpriced usage) is unchanged. - If a provider stop can never be verified, the run stays "recovering accounting": new work waits without a hard-stop incident. Before, the same state paused the company. The hold is visible in admission errors and Costs. - A receipt that arrives after settlement posts a late correction. The original unpriced charge stays in the ledger. - No schema or migration change. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use (shell, file editing, embedded Postgres tests, live restart-chaos harness). ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 14:27:51 -06:00
DottaandPaperclip	0b48bc7ac5	test(connections): isolate scheduled GitHub history scans (#19541 ) ## Thinking Path > - Paperclip lets people manage AI agents and their service connections. > - Connection identity reads must use the authenticated caller's current account grants. > - This integration suite runs the real HTTP routes and bearer authentication against provider fixtures. > - An earlier GitHub App fixture stays active after its test ends. > - Its history worker can run 60 seconds later against the next test's shared provider handler. > - This pull request keeps that unrelated worker inactive in this suite. > - The identity test keeps its original provider-count and authorization assertions. ## Linked Issues or Issue Description What happened? The Anthropic identity integration test reported three provider calls instead of two. The denied request returned `connection_unavailable` correctly. A controlled reproduction traced the third call to GitHub's `/app/hook/config` from an earlier test's webhook recovery worker. Expected behavior The suite must measure requests from the operation under test. Foreign-company, unknown-grant, and revoked-grant requests must not contact the identity provider. Steps to reproduce Run the GitHub App import fixture before the identity test. Initialize the saved GitHub recovery epoch. Advance only the fixture Date by 60,001 ms, then run the actual current recovery check after the two authorized identity reads. The original final count assertion fails with three calls. The extra call's stack runs through `readGitHubAppWebhookConfig`, `recoverGitHubWebhookEndpoint`, and the recovery schedule's `check`. Related implementation: #17666. No duplicate fixture repair was found. ## What Changed - Keep automatic GitHub webhook history work inactive in this connection-management test suite. - Keep real connection setup, HTTP routes, database state, agent API keys, JWT authentication, and all existing identity assertions. - Leave production registration and the dedicated chat-channel recovery tests unchanged. ## Verification - Exact baseline `5d25117ffad69a93fb92026950b9acd9cec08f69`: the unchanged focused identity test passed. - Controlled red on that baseline: the original final `expect(calls).toBe(2)` failed with three. The third URL was `https://api.github.com/app/hook/config`. The revoked identity response stayed unavailable. - Local provider traffic uses fixture responses. No real provider credentials or requests were used. - Full connection-management suite: 90/90 passed on the exact release baseline plus this fixture fix (commit `e33c091438`). Command from `server`: `../node_modules/.bin/vitest run --config vitest.config.ts --fsModuleCache src/__tests__/connection-management.integration.test.ts`. - Rebased onto current master. A fresh GitHub-import plus identity pair passed 2/2 on final head `95ed6147920e48128892744e1a4f220373f7da5b` after removal of the added comments required by review. Greptile reviewed that exact head at 5/5. The addressed comment thread is resolved. Required CI remains pending. Final reviewed head: `95ed6147920e48128892744e1a4f220373f7da5b`. [Required CI](https://github.com/paperclipai/paperclip/actions/runs/42272042111) passed (54 successful checks, three expected skips and the completed neutral Paperclip Review check; none pending or failed). Greptile and Paperclip Review both rated the change 5/5 with no remaining findings. The addressed review thread is resolved. The first attempt's canary runner communication loss and unrelated cold-import timeout are retained as failed evidence; the unchanged cold-import suite passed 4/4 locally, and only failed CI jobs were rerun successfully. ## Risks Low risk. This changes test isolation only. The connection-management suite no longer runs unrelated GitHub history scans. Dedicated chat-channel integration tests still cover those scans. No authorization check, assertion, deadline, or production behavior changes. ## Model Used OpenAI GPT-6 family via Codex, with coding and tool use. The exact model ID and context window are unavailable in this session's metadata. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 14:19:17 -06:00
DottaandClaude Opus 5.5	5a5bfe689f	fix(heartbeat): finalize an accepted native result after a graceful shutdown (#19591 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Native runner runs go through a durable coordinator row (`native_run_finalizations`) that holds the controller's execution lease, the accepted result, and the finalization phase > - A graceful server shutdown (SIGTERM) drains running native runs: it detaches the executor and hands the run to the next server boot > - When SIGTERM lands after `run.result.accepted` / `run.terminal` but before finalization, the drain did not release the execution lease, because it only updated coordinator rows without a result > - Startup restart recovery skips accepted results, and both the finalization reconciler and the orphan reaper wait for an unexpired coordinator lease (20 minutes for a local run). The run stayed `running` and the task stayed `in_progress` with no Retry > - This pull request makes the drain release the execution lease of an accepted result too, so the next boot finalizes the durable result > - The benefit is that a completed task finishes after a restart instead of stalling silently ## Linked Issues or Issue Description Refs #19525, #19530, #19513 What happened? A restart-chaos wave sent SIGTERM to the server 0.75 s after a native OpenCode run emitted `run.terminal` (result accepted, `runTerminalState: succeeded`). The graceful drain listed the run in `restartSuspendedRunIds`. On the next boot, startup recovery claimed only another run. The accepted run stayed `running` for more than 10 minutes. Its task stayed `in_progress` with `interruption.state=working`, and no Retry appeared. Expected behavior After a restart, a run whose result was accepted and whose terminal state is known is finalized as succeeded, and its task is completed. It must never stay silently `running`. Root cause `drainRunningRunsForShutdown` (`server/src/services/heartbeat/recovery.ts`) updated the coordinator only `where result_id is null`. For an accepted result, the update did nothing, so the dead controller's execution lease (`NATIVE_SESSION_EXECUTION_LEASE_TTL_MS`, 20 min) stayed valid. The detached executor does not finalize (by design, it hands off to the next controller). Then: - `claimNativeRestartRecoveries` only claims rows with `result_id is null`. - `reconcileNativeFinalizations` only selects rows whose lease is null or expired. - The orphan reaper skips a native run with pending workspace finalization or an unexpired controller lease, and defers to the lease deadline. The 5-minute stale threshold is not involved. Steps to reproduce See the new regression test: accept a result through `PaperclipControlPlanePort.completeRun` under a held 20-minute execution lease, run the SIGTERM drain, then run startup recovery on a new heartbeat service. ## What Changed - `drainRunningRunsForShutdown` now releases the coordinator execution lease (`leaseExpiresAt = now`) for an accepted result as well. It keeps `recoveryState` unchanged for an accepted result, because the run needs finalization, not a runner reattach. Rows without a result keep the old behavior (`awaiting_runner_reattach`). - A `native-finalizer:` lease on an accepted result is kept. That lease belongs to a finalization that is in progress in the draining process, and the shutdown waits for it. - The drain event for an accepted result now says that the result was released for finalization (`runnerDisposition: awaiting_result_finalization`). - New regression tests in `server/src/__tests__/heartbeat-native-restart-suspension.test.ts`: - An accepted terminal result under a held execution lease, then SIGTERM drain, then next-boot recovery: the run becomes `succeeded`, the coordinator becomes `committed`, the task becomes `done`, there is one result row and one run, and no provider turn starts. - A `native-finalizer:` lease survives the drain unchanged. ## Verification - Red before the fix: `cd server && pnpm exec vitest run src/__tests__/heartbeat-native-restart-suspension.test.ts -t "accepted terminal"` failed with `{ status: 'running', phase: 'workspace_finalizing' }` instead of `{ status: 'succeeded', phase: 'committed' }`. - Green after the fix: the same file passes 5/5. - Related suites pass: `src/services/heartbeat/recovery.test.ts`, `src/services/legacy-controller-lease.test.ts`, `src/services/native-runtime/native-runner-restart-recovery.integration.test.ts` (80 passed, 14 skipped), and `src/__tests__/heartbeat-process-recovery.test.ts` except one case. - The one failing case in `heartbeat-process-recovery.test.ts` ("dispatches interrupted CLI input and cleans up early fixture exits (true)") also fails on unchanged `origin/master`. It is not related to this change. - `server` typecheck (`tsc --noEmit`) passes. - Restart-chaos replay of the original cell: in progress. Results follow in a PR comment. ## Risks - Low to medium. The change only affects the graceful SIGTERM drain of native runs owned by this boot. - A rolling deploy can overlap the old and new servers. The old process detaches the executor before it releases the lease, so it does not finalize the result. A finalizer that already started keeps its `native-finalizer:` lease, so the new server cannot finalize the same result at the same time. - If a `native-finalizer:` lease is in progress and the process exits before it finishes, that run still waits for the 5-minute finalizer lease. This was already the behavior and is out of scope. - No schema, API, or UI change. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, used through Claude Code with tool use (shell, file edit, test runs). Extended reasoning was used. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 14:11:34 -06:00
DottaandPaperclip	124b2cd0ad	fix(server): certify background deadlines for idle tenants (#19001 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Background services own future checks, retries, and recovery work. > - An idle tenant can stop only when a durable wake covers its remaining duties. > - The current alarm catalog covers several domains but leaves these owners blocked. > - A saved future timestamp alone cannot prove that reconstruction or admitted work has settled. > - This pull request binds durable domain snapshots to the actual registered worker revisions. > - The tenant can hand off supported future duties and retain all checks for uncertain or active work. ## Linked Issues or Issue Description No matching public issue was found. Related work: [domain deadline handoff #18656](https://github.com/paperclipai/paperclip/pull/18656), [owned blocker diagnostics #18632](https://github.com/paperclipai/paperclip/pull/18632), and [completed history proof #18646](https://github.com/paperclipai/paperclip/pull/18646). The separate [native idle-session change #18988](https://github.com/paperclipai/paperclip/pull/18988) covers runtime retirement. [Far-future retry re-probes #13534](https://github.com/paperclipai/paperclip/pull/13534) changes recovery responsiveness after an upstream success; this change preserves existing retry decisions. What happened? Future status-card, external-object, skill-sync, watchdog, retry, and quota duties lacked complete owner-bound coverage in the idle deadline proof. A completed quota task could also retain its monitor policy and timestamp and remain blocked after the scheduler had disarmed it. Expected behavior A settled registered owner can hand off its exact supported future deadline. Active work, dirty reconstruction, transaction uncertainty, changed revisions, claims, unknown state, and unsupported duties keep the tenant awake. Completed history retains its data and stops recurring work when the owner has no remaining duty. Steps to reproduce 1. Produce a supported future duty through its normal service. 2. Let the registered worker finish reconstruction and admitted work. 3. Request an owned idle report with the negotiated domain deadline catalog. 4. Check the saved earliest deadline and generation before durable alarm registration. 5. Stop and reconstruct the worker. Check that ordinary admission and claims resume useful work once due. 6. Complete a quota task. Keep its saved policy and timestamp. Check that linked unfinished work still blocks sleep and that settled history does not. Paperclip version or commit The owned patch was qualified at `699c8569fa34338dd75cdc5731381ec1e2efdb45`. Publication head `491fac368ee72eaac8ae105d53e6bf6b305d8a6a` normally merges actual master `039e0dad2c51c80251e1302a19c5566e039c357a`. All 43 owned files are byte-identical. All 29 new upstream paths are preserved. The later normal current-master integration is `8c912e1ba4fc66397a7341ade59373af8075068f`, described separately below. Deployment mode Source checkout with real PostgreSQL 17 and a controlled local Node process. No deployed alarm, tenant suspension, database suspension, or fleet result is claimed. ## What Changed - Add registered-worker reconciliation revisions and invalidate them on stop. - Keep status-card, external-object, and skill deadline interpretation with their owners. - Reuse the watchdog classifier against the supplied database snapshot. - Validate the exact saved retry run and paired queued wake. - Validate legacy quota source bindings and the current assignee or reviewer policy. - Preserve common issue scope, dependency, sibling, and state checks for covered waits. - Compose normalized owner snapshots within the existing bounded deadline proof and recheck revisions after inspection. - Bind actual service and worker instances from server startup. Missing bindings preserve existing blockers. - Retain completed quota policy and timestamps while independently rejecting linked unfinished runs or wakes. - Block unconsumed budget evaluation inputs through the owning queue predicate. Accounting policy and settlement stay unchanged. - Bind completed quota history through fixed-size UUID-array parameters, preserve exact company/task/source associations, and guard external references before the run-ID join. - Batch watchdog subtrees and shared classifier facts without changing the 64-query or two-second proof budget; preserve exact company pairing for first-run evidence. - Extend production-path regression controls and document supported coverage and limits. ## Verification Current integration head: `688abc6cac6016b9fb50e5832f81e2975ed5d49f`. Fresh exact-head checks completed with 58 successes, three intentional skips and one neutral review, including aggregate `ci / verify`. The complete current Greptile review is 5/5. The complete current Ponytail verdict is Ship (4/5), with only optional cleanup of an unused MCP proof input; both prior correctness findings are resolved. All review threads and nested pages were inspected. The normal master merges preserve the registered retry deadline owner and current startup completion log after the same readiness await. The earlier focused evidence below remains tied to its tested source revisions. Current-master integration `8c912e1ba4fc66397a7341ade59373af8075068f`: - Normally merge actual master `46fb933b0155d183ce64a8bdc4950d1b58c926f6`. Every nonoverlapping tree blob retains its owning parent across 14,044 paths. All 42 owner-only files and 15 critical binding/proof/readiness files remain byte-identical. Three automatic overlaps preserve the existing watchdog binding, conservative unavailable-reader fallback and optional actual owner arguments. The two scoped-deadline conflicts retain the entire upstream resource-reservation/mid-pass admission/abort algorithm inside the existing reconciliation revision/running/failure fence, with both parent test sets. - Actual TypeScript 7.0.2 server noEmit passed in 9.434 seconds, normal source declaration layers in 12.102 seconds, and four selected test/helper/config layers extending the actual CI configuration in 3.167 seconds. No compiler option weakening, heap override, donor declarations or SDK replacement stub. - Focused PG0 controls passed: 102 scheduler/proof/owner controls in 16.451 seconds, including all 21 scoped-deadline cases; 121 startup controls; and all 107 unchanged instance-settings route cases in 6.456 seconds. The latter first hit a local prewire-guard listener-registration race. Only the ignored guard was corrected to register the same actual OS-bound server port immediately after native listen returns. Closed, unknown, foreign, offhost and PG destinations remain blocked. Original failure evidence is retained; product code and assertions were unchanged. - The exact current UI policy command now reports zero findings (0.362 seconds); its seven scanner controls pass. The earlier dependent PR used an old scanner/source baseline under a newly enforcing trusted workflow and reported 397 findings. Carrying the genuine current master resolves that mismatch without editing unrelated UI sites or relaxing the gate. - Every meaningful worker checked its executing PID/thread/preload hash and matching receipt. PG0 controls opened no PostgreSQL sockets; all owned processes exited. Original dependency-cache/DNS, ignored-harness syntax/type and guard failures remain separate from product results. - This is fresh semantic and focused runtime qualification, not a local whole-repository or current-head PostgreSQL acceptance rerun. The upstream workspace-busy parked-retry/terminal-holder notification path remains unchanged by this merge and needs its separate current-graph producer check. Historical six-case/scale results below stay tied to their measured sources. A matching controlled default-cadence follow-up is in #19191; no fresh current-head two-cycle or deployed suspension claim follows from this merge. - All 59 exact-head check contexts now pass (55 success, one neutral, three intentional skips), plus the legacy security status. The complete edited Greptile review is 5/5 for this exact commit, with all root/nested pages inspected and no unresolved threads. Ponytail stopped without completing its new turn; it supplied no new verdict or finding. Prior actual findings remain addressed. Earlier scores/checks below belong to their explicit earlier commits. On frozen clean merged source `699c8569fa`: - Real PostgreSQL acceptance: 6/6 passed,83.483 seconds wall. Actual retry producer creates the exact saved run/wake; stopped and reconstructed owner resumes through current admission, one durable start, real local process output and normal task completion. Includes cancellation, drain arrival, retained quota history with queued-run/wake negatives, watchdog review settlement and budget-input empty/pending/consumed control using the actual owner. - Affected coordinator, deadline, scoped-deadline, recovery and composed-proof controls: 78/78 in five files passed,32.114 seconds wall, worker 1/no file parallelism. - `pnpm exec tsc --noEmit -p server/tsconfig.json`: passed,17.467 seconds. Native TypeScript 7.0.2 and Node 24.22.2. - PostgreSQL tests use one dedicated test database, matching fixture/admin URLs, pool limit four, one worker, and no embedded fallback. Every run recorded unchanged source hashes and a clean checkout. - The eight-second real-clock retry is a domain execution check. It is below the Cloud minimum wake lead and is not a positive accepted hosted sleep cycle. - The completed quota-history product failure and original fixture/protocol failures were preserved; the owning-reader fix passed a real PostgreSQL red/green regression before the fresh merged six-case pass. - After runtime completion, the dedicated database had zero other clients and no owned runtime processes. All three qualification process PIDs exited. The coordinator independently audits and drops the database. - Comment-only cleanup has equal strict TypeScript parser ASTs across 15 files. The final lexical audit finds no added comments relative to actual master. `git diff --check` passed. - Earlier retry fixtures omitted responsible-user membership, advanced JavaScript time past the real PostgreSQL lease clock, or produced stdout without normal task completion. Corrected fixtures use real membership, an eight-second real deadline, and ordinary task comment/completion. The last case exposed legitimate disposition repair rather than duplicate retry promotion. - The quota and watchdog controls first ran before admitted work or the existing 30-second reconstruction duty settled. The stronger tests require an interim blocked report and then exact settled-owner proof. Production deadlines were not changed. The separate completed-quota-history defect was fixed in its reader and verified with linked unfinished-run and wake controls. - The status/object and skill owners reviewed the actual final startup, route and post-backup bindings. Their domain source evidence carries where token-equivalent. The master coordinator coalescing change was reviewed separately. Historical domain PostgreSQL results are not presented as fresh full-suite results on this head. The later master merge adds retry predecessor cleanup checks under the existing issue claim lock. The source audit confirms unchanged saved deadlines and run/wake identities. The three actual retry lifecycle/cancellation/drain cases passed on the batched candidate with this parent (31.817 seconds). Those local dependency controls preceded the current exact-head remote run. Native backup preservation and TestOps changes from master are retained. No pass is inferred for changed dependency behavior solely from unchanged owned files. Review repair published as `643f1e7400`: - The original reader returned `unknown` after seven actual watchdog reviews settled across seven companies. The unchanged composed-report regression passed after batching; the final reader passed in 61.841 seconds, including active queued-work and stopped-owner rejection. - A separate two-company red/green control rejects terminal-run context from another company as first-run grace evidence. Shared facts are serialized once, with per-root IDs and review summaries. - The affected recovery, proof and diagnostic unit controls passed 47/47 (14.575 seconds). Startup owner-contract fixtures passed 121/121 (10.589 seconds). - Diagnostic PostgreSQL controls passed7/7 (12.406 seconds), including the normal budget-input producer and its exact blocker reason. Initial empty-state setup uses the same cleanup as per-case teardown when a dedicated test database is reused. - Actual repository `pnpm exec tsc --noEmit -p server/tsconfig.json` passed 12.710 seconds with no heap override. Final source has no added code comments; whitespace checks pass. - The original query-budget failure, two-company discriminator, initial cross-suite fixture failure and an interrupted incorrect local-driver selector are retained separately. The interrupted run is not a pass. Independent cleanup found zero owned clients or runtime processes. Quota-scale repair published as `6696a37f8a`: - The exact 10,000-row completed-history case reproduced `MAX_PARAMETERS_EXCEEDED`:90,002 bound parameters exceeded the driver's 65,534 limit, and the composed report returned `unknown` despite settled owners. Original failure and actual driver diagnostics are retained. - Fixed pending-work queries use three typed UUID-array parameters plus the result limit, keeping exact company/issue/source tuple bindings. The primary run-ID column stays uncast; a guarded CASE validates the raw external reference before conversion. The 10,000-row cap and 64-query/two-second proof budget are unchanged. - The same 10,000-row composed-report case passed 18.860 seconds, with unfinished run/wake rejection, malformed-reference safety and conservative 10,001-row overflow. Its fixture derives retained history from the actual quota producer and normal task completion, expands valid persisted issue/source pairs, then starts the actual registered owners. - The existing actual quota producer case passed 14.187 seconds, including exact future deadline, retained completion history, unfinished work, real foreign-company source references and noncanonical/malformed references. - Affected owner/proof controls passed 22/22 (26.444 seconds). Production server noEmit passed 16.039 seconds. - CI on 643 identified TS7034/TS7005 in the settled-watchdog fixture's initially untyped collection. It is now explicitly typed. A separate affected-entry semantic check extending the repository's actual `server/src/__tests__/tsconfig.json` passed 15.093 seconds: six affected tests, two owner files, ambient declarations and current imported source, with unchanged strict/module/lib/path settings. The production server config excludes that test directory and is not presented as test-layer proof. No custom suppression or replacement SDK stub was added. - All runs were sequential with pool max 4, worker 1, source/config/command hashes and cleared inherited provider/LIVE/other-database variables. Cleanup found zero owned database clients or local runtime processes. At prior head `6696a37f8a5718cd0f79054969a0084462d41d47`, all required CI gates pass. The exact-head inventory has 55 successes, one neutral review and three intentional skips. Greptile is 5/5. The complete latest Ponytail verdict confirms the quota finding is fixed; its only remaining note is optional unused MCP proof state, not a behavior defect. All three review threads are resolved. The earlier 643 result remains historical. The repository-wide typecheck, tests, build and browser gates passed in the prior `6696a37f8a5718cd0f79054969a0084462d41d47` CI run. The new `8c912e1ba4fc66397a7341ade59373af8075068f` full CI also passes. They were not run as a full local workspace command in the focused PostgreSQL lease. Local evidence above uses bounded external databases and does not claim embedded-cluster, live-provider or deployed-alarm qualification. ## Risks - An unsupported or uncertain owner state must remain blocked. An optimistic exemption could lose work while compute is stopped. - Changes during reconstruction or inspection must invalidate the proof. A memory-only timer is not durable alarm coverage. - Current authorization, budgets, cancellation, accounting, and resource ownership still apply on dispatch. The proof does not grant execution authority. - The supported writer model uses in-process transaction signals and startup/resume reconstruction. Independent SQL writers and other replicas need an explicit wake contract. - Native, review, coalesced, and taskless retry variants remain conservative until their owner cases qualify. Reviewer quota and hosted alarm qualification remain open. - The shared catalog remains opt-in. Legacy callers and missing owner bindings retain their old blockers. ## Model Used OpenAI `gpt-6.1-sol` was the configured model for the delegated service implementations and reviews. OpenAI Codex, based on GPT-6, performed managing review and integration. Both used repository tools and code execution. The exact served backend snapshots and context-window limits were not exposed. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change and contains no internal Paperclip ticket id - [x] I have run tests locally and they pass on the final merged head - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green — exact current-head run passes - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups — complete current-head review and all paginated threads inspected - [ ] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 13:48:55 -06:00
DottaandClaude Opus 5.5	694076245f	fix(status-cards): inline query guidance in the compile task instead of an unprovisioned skill (#19554 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Status cards let the board watch a slice of company work with a prose prompt > - A new card starts a compile task. The summarizer agent compiles the prompt into search queries and writes the first summary. > - The compile task told the agent to use the bundled `status-card-query` skill. Nothing gives that skill to the built-in Summarizer or to a per-card override agent. > - The agent saw no assigned skills, got `runner_skill_not_assigned`, and then searched the filesystem until the run deadline. It never wrote the card query. > - This pull request puts the query guidance directly in the compile task, so the task names only things the run has. > - The benefit is that any summarizer agent, on any model, can compile and save the card query without a missing dependency. ## Linked Issues or Issue Description What happened? Creating a status card with a summarizer on a smaller model (for example an OpenCode agent on a DeepSeek flash model) never finished. The compile run reported no assigned skills, then spent its full budget looking for `status-card-query`. No `PUT /api/status-cards/{id}/query` request was sent, and the run hit its deadline. Stronger models sometimes worked around the gap. Expected behavior The compile task contains everything the run needs. The agent writes the query, reads the compiled scope, and writes the first summary. Steps to reproduce 1. Enable status cards and configure a summarizer agent that does not have the `status-card-query` skill (the default Summarizer only installs `summarize-status`). 2. Create a status card. 3. Watch the compile run: it looks for the skill and does not write the query. ## What Changed - `server/src/services/status-cards.ts`: the compile task no longer says "Use the bundled `status-card-query` skill". It now inlines the query format (fields, allowed status and priority values from shared constants, limits), the project and label lookup endpoints, an example query body, and an explicit `GET /api/status-cards/{id}/dry-run` step before the summary write. - The example body uses a `text` fence, so the existing parser still reads the `json` generation payload. - New `server/src/__tests__/status-card-compile-prompt.test.ts`: asserts that the compile task names no skill, contains the full query format and write-back sequence, has an example that passes `companySearchQuerySchema`, and keeps a parseable generation payload. ## Verification - `cd server && npx vitest run src/__tests__/status-card-compile-prompt.test.ts src/__tests__/status-cards.test.ts src/__tests__/status-card-evaluation-work.test.ts`: 3 files, 74 tests pass. - `npx tsc --noEmit -p server`: no errors in the changed files. - `pnpm verify` selected 2224 test files. Most failed in this fresh worktree because `@paperclipai/plugin-sdk` was not built yet (`Failed to resolve entry for package "@paperclipai/plugin-sdk"`). After a build of that package, the status-card files above pass. Required CI supplies the broad coverage. ## Risks - Low risk. The change is prompt text only. The write-back routes, validators, and authorization do not change. - The compile prompt is longer. The added text is a small, bounded block. - The `status-card-query` catalog skill still exists for agents that author cards. The two texts can drift. The new test pins the prompt contract. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), via Claude Code with tool use and code execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 13:48:45 -06:00
TonioandPaperclip	c650f36007	fix(ui): refine agent profiles, identity editing, and list actions (#19535 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Agent pages expose identity files, reporting relationships, and runtime settings. > - Several controls had inconsistent spacing, weak navigation, or misleading save feedback. > - Empty task-row detail slots also reduced the title width before truncation. > - This pull request fixes those behaviors and aligns the reviewed UI details. > - The result is clearer agent editing and more readable task and agent lists. ## Linked Issues or Issue Description What happened? Agent headers and file controls had inconsistent layout. The Identity footer appeared without edits and did not provide clear save confirmation. File names could wrap or omit the Markdown suffix. Task titles truncated while empty detail slots occupied the row. Expected behavior Keep headers and actions aligned. Reveal the Identity footer only for changes or save feedback. Use valid Markdown file names and actual file dates. Let titles use available space. Steps to reproduce 1. Open an agent overview, then its Identity tab. 2. Edit, undo, save, copy, and add an instruction file. 3. Resize the file panel and inspect file names and date labels. 4. Hover an agent row and inspect the Chat action and tooltip. 5. Open a task list with no recovery or external details and inspect title width. Related: #18716 is the prior UI bundle. #16934 and #14483 concern role editing; this change does not add a role picker. ## What Changed - Refine the character header, inline title and name editing, crown position, runtime shortcut, and Chat/New task actions. - Align profile metadata and manager selection; hide empty direct reports. - Refine Identity panels, Markdown filename validation, file-row hover, truncation, Paper Mono typography, and copy feedback. - Add a viewport-bound save footer with dirty, saving, and saved states. Keep undo/discard separate from successful saves. - Return recorded creation or modification dates for instruction files. Reuse the active file transaction for metadata reads to avoid exhausting the small connection pool. - Reveal the agent-row Chat icon and label on hover, with a standard delayed tooltip. - Collapse empty task-row details so titles can use the available width. ## Verification - Browser checks covered the local overview and Identity flows, narrow file panels, successful file creation/update, undo/discard, header navigation, row hover, tooltip delayed-open state, and full task titles. - Focused instruction, save-footer, date, font-asset, and task-row tests passed during development. After review, 55 file/instruction tests passed, including cold restoration with recorded and unknown dates; server typecheck and 18 immediate-save response tests also passed. - Full `pnpm -r typecheck` and `pnpm build` passed after rebase onto current master. The complete `pnpm test:run` attempt hit a 15-second timeout in the existing chat Retry HTTP integration test; that case passed in isolation (1 test passed). A second full attempt hit timeouts in two other chat integration cases (GitHub webhook draining and delivery retry budget); both passed in isolation without increasing timeouts. All 55 remote checks passed on `f31405a730`, including the full test matrix, build, typecheck, and final verification gate. - `pnpm check:token-gates`, `node scripts/check-maps.mjs`, and `git diff --check` passed on the submitted revision. ## Risks - File dates are optional. Managed files require recorded dates because restored projections have synthetic filesystem timestamps. External files can use modification time; unknown dates are omitted. - Save-footer timing depends on animation completion; reduced-motion and failed-save paths have tests. - Global Paper Mono typography can change text widths. Browser checks covered the edited surfaces. - No schema migration or new execution permissions are introduced. ## Model Used OpenAI Codex, GPT-6, with repository tools, code execution, and browser verification. The exact serving model ID and context-window size are not exposed in this session. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes:` / `Closes:` / `Refs:` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub references) - [x] My branch name describes the change and contains no internal Paperclip ticket id or instance-derived details - [x] I have run the relevant tests locally and they pass; full-run timeout limitations and successful isolated reruns are documented above - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>	2026-11-06 11:14:56 -08:00
DottaandClaude Opus 5.5	705c3e955d	fix(chat): show one onboarding question card when onboarding moves to agent chat (#19512 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - New users start with an onboarding first task. It has an opening question card and, after "Interview me", an interview question card. > - With the coordinator chat enabled, the agent's chat with the user mirrors pending questions from that agent's direct tasks. The onboarding task is a direct task. > - A tester moved onboarding from the task into the chat. The agent asked its interview in the chat. The chat then showed the chat card and the mirrored onboarding-task card. Both were answerable. > - This pull request expires the agent's pending onboarding-task questions when the agent asks a question in its chat with that user. > - The benefit is one interview card, answerable once, wherever onboarding runs. ## Linked Issues or Issue Description What happened? Onboarding testing on Paperclip Cloud staging (Codex API path). The steps were: enable the experimental chat and coordinator features, start the "Interview me" onboarding task, switch to the agent chat, and ask the chat to run the onboarding there. The chat agent started the interview, and a duplicate question card appeared. Root cause `conversationDecisionReceiptsService.listPendingQuestions` projects pending questions from direct tasks (created by the chat user, assigned to the chat agent) into the chat. The onboarding first task matches. The PAP-437 sibling-supersede rule in `issueThreadInteractionService.create` only works on the same issue, and it is off for chat questions. So the task's pending interview card and the new chat card both stayed pending and visible in the chat. Expected behavior One interview question card in the chat, answerable once. Steps to reproduce 1. Enable the experimental Agent Chat and coordinator features. 2. Create the onboarding first task and choose "Interview me" on its opening card. 3. Open the chat with the same agent and ask it to run the onboarding there instead of in the task. 4. The agent asks its interview in the chat. The chat shows two pending question cards: its own card and the mirrored onboarding-task card. Paperclip version or commit `master` at `819420a60a` (Paperclip Cloud staging build at the time of the test). Deployment mode Paperclip Cloud (staging). The same logic runs in self-hosted mode. ## What Changed - `server/src/services/issue-thread-interactions.ts`: when the chat agent creates an `ask_user_questions` card in its chat with a user, the server expires that agent's pending `ask_user_questions` cards on the onboarding first task that the same user created and assigned to the agent. The result is the existing `superseded_by_newer_interaction` shape and points at the chat card. Pending chat publications for those cards are cancelled and terminal publications are queued, the same as the existing sibling supersede. - The rule does not change questions on other direct tasks, handoff tasks, or other chats. Confirmations and plan approvals on the onboarding task are not changed. - `server/src/__tests__/coordinator-handoff.test.ts`: a new red/green test for the reported flow, and a control test that an ordinary direct-task question stays pending when the agent asks something else in chat. - `server/src/onboarding-assets/first-task/README.md`: one paragraph that documents the behavior. ## Verification Red on `master` (`819420a60a`) with only the new tests added: ``` $ cd server && npx vitest run src/__tests__/coordinator-handoff.test.ts -t "onboarding interview\|ordinary direct task" × moves a pending onboarding interview into the agent chat instead of showing two question cards AssertionError: expected [ …(2) ] to deeply equal [ Array(1) ] [ "26b1e6ae-e1e5-4c26-8475-ad7ad93a7754", + "2364fd68-a40f-42be-854d-89210bfc5811", ] ❯ src/__tests__/coordinator-handoff.test.ts:231:53 Tests 1 failed \| 1 passed \| 23 skipped (25) ``` The chat showed its own interview card and the mirrored onboarding-task interview card. Green after the fix: ``` $ cd server && npx vitest run src/__tests__/coordinator-handoff.test.ts Test Files 1 passed (1) Tests 25 passed (25) ``` Neighbouring suites: ``` $ npx vitest run src/__tests__/issue-thread-interactions-service.test.ts src/__tests__/issue-onboarding-first-task-routes.test.ts src/__tests__/chat-interaction-publications.test.ts src/__tests__/coordinator-handoff.test.ts Test Files 4 passed (4) Tests 183 passed (183) ``` `tsc --noEmit -p server` shows no errors in the changed files. Other errors in that local run come from the unbuilt `@paperclipai/paperclip-runner` package. CI typecheck covers the full build. Not done: a live browser repro. Staging has no Codex company and no agent chat with this flow. The evidence is the tester's notes and the service-level test. ## Risks - If a user ignores the onboarding card and the agent asks an unrelated question in its chat, the onboarding task's pending question card expires. The onboarding task stays open, and a message on the task still wakes the agent. The chat already showed both cards to that user, so this keeps one live ask from the agent. - Race: if the onboarding task run posts its interview card after the chat card, both stay pending. Agent Chat questions are intentionally not expired by other cards, so this PR does not reverse the rule. - No schema, API, or UI change. ## Model Used - Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with tool use and code execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 13:10:45 -06:00
DottaandClaude Opus 5.5	fca1f1be4f	fix(attention): keep replaced confirmations out of Needs you after the newest is settled (#19520 ) ## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Needs you is the board inbox. It lists the decisions that wait for a person. > - Agents often ask for the same confirmation again on one task. Needs you shows only the newest pending confirmation for each task. > - When a person approved that newest card (for example in chat), or the card expired, the older card that it replaced came back into Needs you. > - Manual onboarding tests reported this: decision cards that were already approved in chat still showed in Needs you. > - This pull request makes the newest answered or expired confirmation also hide the older pending confirmations on the same task. > - The benefit is that Needs you does not show stale cards after a person answers the current one. ## Linked Issues or Issue Description What happened? A person approved a decision card in chat. Needs you still showed a decision card for the same task. Expired cards also caused an older card to show again. Steps to reproduce 1. An agent creates a confirmation on a task. Then it creates a second confirmation on the same task. 2. Needs you shows only the second card. 3. Approve the second card in chat (or let it expire). 4. The first card now shows in Needs you. Expected behavior Needs you shows no card for the task, because the newest ask is settled. Root cause `collapsePendingConfirmationsToNewest` compared only pending rows. Chat cards use the same `issue_thread_interactions` row and the same accept service, so there is no duplicate record. The server already drops answered rows. The problem was the older pending row that the collapse hid before. ## What Changed - `server/src/services/attention-source-eligibility.ts`: the collapse takes the newest settled (accepted, rejected, answered, expired) ordinary confirmation for each issue. Older pending ordinary confirmations stay hidden. A new `loadNewestSettledConfirmationAt` helper reads that value with one grouped query. - Cancelled and failed replacements do not settle the ask, so the older card returns (the existing retention test keeps this behavior). - Tool-action confirmations and quality-review confirmations are excluded from the settled check. A settled tool action never hides a different pending tool action. - `server/src/services/attention.ts` and `server/src/services/attention-retention.ts` use the new input, so the feed and the retention worker agree. - Review follow-up: the summary and task-index payload keeps `toolAction` and `secretProposal`. A settled card only hides older asks addressed to the same recipient. Tool actions, secret proposals and app connections (`connectionAuthorization`) form one governed set. They match the set the other confirmation readers use. A settled governed card never settles an ordinary ask, and a pending governed card is never hidden. - `server/src/__tests__/attention-service.test.ts`: new tests for "approved from chat", "rejected from chat", "expired", a re-ask after a settled card, another recipient, a settled secret proposal, tool actions in `list()`, and tool actions in `summary()`/`taskIndex()`. The tests use two agents, because the service already expires a same-agent re-ask at creation. ## Verification Red on `master` (`819420a60a`), before the code change: ``` RUN v5.0.3 /Users/dotta/paperclipai/paperclip/.claude/worktrees/fix-n10-stale-needs-you-cards-47c895/server ❯ src/__tests__/attention-service.test.ts (36 tests \| 2 failed \| 34 skipped) 14165ms ❯ attention service (36) × does not resurface an older confirmation once the newest one is approved from chat 3502ms × does not resurface an older confirmation once the newest one is expired 2097ms ⎯⎯⎯⎯⎯⎯⎯ Failed Tests 2 ⎯⎯⎯⎯⎯⎯⎯ FAIL src/__tests__/attention-service.test.ts > attention service > does not resurface an older confirmation once the newest one is approved from chat AssertionError: expected [ Array(1) ] to deeply equal [] - Expected + Received - [] + [ + "a1f2f085-e4b2-42f1-ac3c-71d85f6e164c", + ] ❯ src/__tests__/attention-service.test.ts:1609:32 1607\| } 1608\| 1609\| expect(await pendingIds()).toEqual([]); \| ^ 1610\| }); 1611\| ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/2]⎯ FAIL src/__tests__/attention-service.test.ts > attention service > does not resurface an older confirmation once the newest one is expired AssertionError: expected [ Array(1) ] to deeply equal [] - Expected + Received - [] + [ + "8c8d01ad-1084-47b2-927a-24d18529a9a6", + ] ``` Green after the change (same tests): ``` $ npx vitest run src/__tests__/attention-service.test.ts -t "resurface an older confirmation\|tool-action approval" ✓ does not resurface an older confirmation once the newest one is approved from chat ✓ does not resurface an older confirmation once the newest one is expired ✓ keeps a pending tool-action approval visible after a newer tool action is settled ``` Review follow-up red (on `ebe8e0c0ae`, before the follow-up code change): ``` × keeps a pending tool-action approval in the summary and task index after a newer ordinary confirmation is settled × does not let another recipient's settled confirmation hide this recipient's pending one × does not let a settled secret proposal hide an older ordinary confirmation AssertionError: expected [] to include '4c636383-…' AssertionError: expected [] to deeply equal [ Array(1) ] ``` All three pass after `9245bd7b17`. Second review round red (before `a36923c7fc`): ``` × does not let a settled app connection hide an older ordinary confirmation AssertionError: expected [] to deeply equal [ Array(1) ] ``` It passes after `a36923c7fc`. Neighbouring suites (attention-service, attention-retention-work, issue-review-attention, issue-blocker-attention): ``` Tests 125 passed (125) Test Files 4 passed (4) ``` Staging check (read-only): the ONB staging company had one pending Needs you item, on a task with no newer card. Staging has no Codex company, so I could not reproduce the Codex path there. ## Risks - Low risk. The change is read-only and adds one grouped query when pending confirmations exist. - Older ordinary confirmations from the same agent were already expired at creation. This change affects asks from different agents or users on the same task. - The service already expires an older ordinary confirmation from the same agent at creation. This change affects asks from different agents or users on the same task. - Behavior change: an older ordinary confirmation that is still pending stays out of Needs you after a newer one is answered or expired. It stays visible in the task thread. - Not changed: the live-update listener does not match `issue.interaction.withdrawn` or `issue.interaction_cancelled`. Those cards can stay in the client cache until the next refetch. ## Model Used - Anthropic Claude Opus 5.5 (`claude-opus-5-5`) in Claude Code, with tool use and code execution. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>	2026-11-06 13:10:26 -06:00

1 2 3 4 5 ...