mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-10 12:37:55 +01:00
f0ddccff71f3bbda975cac8a86053a67c5d554f3
7726
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 DottaandPaperclip
|
f0ff466aaf |
fix(testops): wait for complete attachment evidence (#18509)
## Thinking Path > - Paperclip manages AI agents and their work. > - Product E2E evaluations check real user requests and their saved results. > - The TestOps reporter also verifies the uploaded result and attachment evidence. > - One failed attachment-row snapshot stopped the reader before its existing observation allowance. > - A later read with the same protected credential reference verified the original result. The original failed subfield remains unknown. > - This pull request handles attachment readiness within the existing bound while preserving the complete final evidence predicate. > - The reporter can wait for complete evidence without accepting missing, inherited, or incorrect files. ## Linked Issues or Issue Description Refs #18477 and #18470. **What happened?** The chat case passed its ten checks and cleanup in [run 41217990335](https://github.com/paperclipai/paperclip/actions/runs/41217990335), but the final reader failed at `attachment_page / attachment_rows`. The unchanged reader later verified that same retained result through the protected reporting identity in [read-only run 41268693639](https://github.com/paperclipai/paperclip/actions/runs/41268693639). This proves the observed snapshot differs. It does not identify which original row subfield failed or prove an upload race. The reader already permits twelve polls and 120 seconds for complete result and attachment metadata. It nevertheless rejects `missed === true` immediately. That flag cannot satisfy complete evidence, but it can be observed again within the existing allowance. **Expected behavior** Wait within the original limit for attachment evidence to become complete. Accept only the same full source, case, status, cleanup, attachment-name, media-type, byte-count, and non-missed predicate. Reject malformed IDs and inherited rows immediately. Permanent missing or incorrect evidence must still fail. **Steps to reproduce** 1. Call the real verifier with otherwise valid responses and one `missed: true` attachment. 2. Return the complete attachment on the second poll. 3. The old reader fails on the first snapshot. The regression now reaches the original complete acceptance predicate on the second snapshot. **Paperclip version/commit** Base `987e0932bf66e703750cd04698b3c7e8993580a7`. **Deployment mode** TestOps reporting observer after a Product E2E evaluation. This change starts no model or app and changes no feature oracle. ## What Changed - Move the strict `missed === true` condition from immediate structural rejection into bounded readiness. - Keep page shape, counts, unique positive IDs, inherited-row rejection, and every source and case guard hard. - Require the same complete final signature and status before success. Keep twelve polls, five-second intervals, the 120-second deadline, and request limits. - Retain closed per-poll counts and readiness booleans on verification and unverified failure. Clone failure observations so callers cannot change retained diagnostics. - Add seven focused regression and negative controls. Sync one existing consumer test with the precise new failure phase and predicate. ## Verification - Genuine parent RED: the real missed-then-complete verifier test failed at the original attachment-row guard. - API owner tests: 60 passed. Direct CLI, exporter, reporter, and retained-reader consumer tests: 31 passed. Node 26.4.0. - Permanent missing evidence reaches twelve polls and eleven waits, then fails. An elapsed 120-second deadline stops after the first page. Malformed IDs and inherited rows still fail immediately. Incorrect byte counts never pass. Closed diagnostics exclude private fields and return detached observation copies. - Three Node syntax checks and `git diff --check` passed. - Two review reproductions failed on the first PR head: a null row lost its controlled invalid-ID error, and a final-GET deadline crossing lost its exact readiness predicate. The corrected head preserves both diagnostics. Null rows still fail immediately; no deadline or acceptance condition changed. - No local app, database, provider, native build, or new evaluation ran. The MJS change needs no TypeScript emission. - [Normal CI 41269715187](https://github.com/paperclipai/paperclip/actions/runs/41269715187) passed on exact head `4e77c1220b0ba81e971591c1c7d28f9064e992cc`: 55 successes, three skipped checks, and one neutral review. Both required checks passed. Greptile is 5/5 and review threads are resolved. - The controlled normal TestOps-launched confirmation is pending merge. The family unit roster runs the local controls; normal PR CI uses separate runner suites, so CI green alone does not claim that roster ran there. ## Risks - A missed file may remain missing. The observer fails if complete evidence does not arrive inside the original allowance. - This is readiness handling, not a claim that `missed` caused the historical failure. Prior passed case grades and failed reporting workflows remain retained. - The previous fixed retained diagnostic intentionally pins the original helper. Its completed receipt remains valid; after this helper changes, that old diagnostic fails closed instead of silently replaying a different observer. ## Model Used OpenAI `gpt-6.1-sol` through Codex, with reasoning, code editing, shell tools, and independent source review. The host 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>canary/v2026.1101.0-canary.3 |
||
|
DottaandPaperclip
|
9dbbfa4fb2 |
fix(qa): wait for task document controls after reload (#18503)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Task documents keep saved work and its revision history with each task. > - The browser document journey must reopen the saved task and read the revised document after reload. > - Reload can finish before the task controls mount. > - The helper checked visibility once and could wait for a tab launcher inside a panel it had never opened. > - This pull request waits for the existing usable controls before it selects the normal UI path. > - The saved content, revision, ownership and no-agent-work checks stay unchanged. ## Linked Issues or Issue Description Refs: #18356. **What happened?** The document case in QA run 41253428446 passed document creation and revision-history checks, then failed during reopen or readback. The preserved screen shows the revised document and revision 2 in chat. The original result has the generic `journey_operation_failed` code. Its exception stack was intentionally excluded, so the exact historical failing instruction is not known. A separate Chromium control reproduces a concrete navigation race in the same helper. If task controls mount 200 ms after reload, the helper can select `Open a new tab` before it opens the properties panel. That control never becomes visible. The same race affects a delayed classic document editor or an existing Artifacts tab. **Expected behavior** The helper waits for the current task's normal authoring or panel controls. It then opens Artifacts through the same visible controls. A task without document authoring must still fail. A stale body, stale title or damaged revision history must still fail. **Steps to reproduce** 1. Open a task from the Backlog list and reload it. 2. Delay mounting the task controls by 200 ms. 3. Run the unchanged document reopen helper with its properties panel closed. 4. Observe that it waits for `Open a new tab` while `Show properties` is visible. **Paperclip version or commit** Frozen source base: `b889ac39b7fbed558c5f4cfbbc2b73c4ca916810`. No duplicate document reopen PR was found. The related document feature is #18356. ## What Changed - Wait for an existing authoring, properties, Artifacts or launcher control before choosing the navigation path. - Wait for usable panel content after opening properties. - Add five credential-free Chromium controls for delayed collapsed, Artifacts, launcher and classic layouts, plus missing-authoring rejection. - Keep the existing Backlog navigation test compatible with the new readiness probes. - Register the browser regression in the existing support suite. No product code, model, budget, run cap or campaign deadline changed. The existing document contract needs no documentation change. ## Verification - Original failure is preserved. The [sanitized QA report](https://pages.paperclip.ing/evals-qa-smoke-41253428446-1/) records both partial checks and the failed journey. This PR does not change that grade. - Before the fix, the real Chromium control failed for delayed collapsed, Artifacts and classic controls. The launcher path and missing-authoring rejection passed. A 750 ms action timeout bounded this diagnostic; final tests use the existing suite timeouts. - Final Chromium: `node node_modules/@playwright/test/cli.js test --config tests/runner-e2e/playwright-support.config.ts task-document-navigation.spec.ts` — 5 passed. - Document units: `node node_modules/vitest/vitest.mjs run --config tests/runner-e2e/vitest.config.ts tests/runner-e2e/product-journeys/task-document-flow.test.ts tests/runner-e2e/product-journeys/task-document-evidence.test.ts --maxWorkers 1` — 23 passed. This includes 19 negative content, ownership and revision controls. - Scoped strict TypeScript, map checks, token gates and diff checks passed. - Node 24 normal QA import preflight passed with network disabled. - The entire `taskDocumentPersistence` function and the canonical evidence module are byte-identical to the base. The existing exact title/body checks, independent API reads, revision pointers and no-agent-work checks remain in force. - Repository-wide build, typecheck and tests were not run locally for this four-file harness change. [Canonical CI 41255897484](https://github.com/paperclipai/paperclip/actions/runs/41255897484) passed on exact head `ffc04500cb0728f6fbc8b3bc2e704c547015641d`: all 59 checks are terminal (55 success, 3 skipped, 1 neutral), with no failed or pending checks. Greptile and Paperclip each gave this head 5/5 with no findings. All review threads were checked; none are open and there is no next page. The branch is CLEAN and MERGEABLE. - Actual QA measurement now passes in [workflow 41266385118](https://github.com/paperclipai/paperclip-cloud/actions/runs/41266385118), campaign `journey-1793502283113-8f658807`. The measured harness `5c0830f693da2ccef062612146b76629a0177049` contains the three exact reviewed document source/test blobs and unchanged canonical evidence module from this head. The target was the QA cloud instance on App `756599e4ea069c9338c320fe8846523d3dfb92a0`. All four canonical checks and all five phases passed; the manifest is COMPLETE with evidence cleanup complete and nothing pending. A positive no-agent-work check establishes zero native work despite omitted raw run IDs. A separate read-only verifier reread the exact original final grade and confirmed the retained-result validator. Raw document snapshots were not retained; content, revision and author proof comes from the executed unchanged exact UI/API predicates. The full 14-case launch remains failed because hiring failed separately; no aggregate pass is claimed. ## Risks The historical exception was intentionally sanitized. The local control proves this helper race, but cannot prove which instruction failed in the prior QA attempt. The current staging measurement passes this document case; the exact historical failing instruction remains unproven. The new readiness waits use the existing assertion timeout and campaign deadline. They do not accept missing controls or weaken document evidence. ## Model Used OpenAI Codex, based on GPT-6, with reasoning, code execution and browser automation. The exact serving model revision 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 `#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 Live qualification: the document case is verified on QA with complete cleanup. Historical failed results remain unchanged. Co-authored-by: Paperclip <noreply@paperclip.ing> |
||
|
DottaandPaperclip
|
987e0932bf |
test(testops): inspect a fixed retained chat result (#18477)
## Thinking Path > - Paperclip manages AI agents and their work. > - The chat feature evaluation checks a real task request on AWS. > - The case passed, but its final TestOps reader rejected attachment metadata. > - The later browser view passes the same check. It uses a different identity and time. > - Another model run cannot explain that difference. > - This pull request adds a fixed read-only check of the retained measurement with the reporting API identity. > - The check separates its own executor from the original measurement and preserves the original grade. ## Linked Issues or Issue Description Refs #18470 and #18462. **What happened?** The real chat case passed all ten checks and cleanup in [run 41217990335](https://github.com/paperclipai/paperclip/actions/runs/41217990335). Final verification failed at `attachment_page / attachment_rows`. The reader had already accepted the source, run, case, status, page counts, and unique attachment IDs. Later interactive metadata had valid IDs and false attachment flags. That later view does not explain the original API caller failure. **Expected behavior** A read-only operation can inspect the same retained result with the existing protected reporting identity. It must report the exact current attachment flags without running a model, uploading results, or changing the original grade. **Steps to reproduce** 1. Read the closed failure from the linked evaluation run. 2. After this workflow is reviewed and deployed, dispatch its fixed no-input operation on master. 3. Inspect its closed receipt. It records both the new executor and the original measurement. **Paperclip version/commit** The retained measurement used `e40de0c9e88210dc476dcf651461c0a8d1b98a59`. **Deployment mode** GitHub Actions metadata job with the existing protected reporting environment. No app, database, provider, or AWS worker starts. ## What Changed - Add one manual no-input workflow for the fixed retained measurement. - Admit the live executor and the original run separately. Verify exact selection and sanitized export artifact hashes before TestOps authentication. - Read only the five existing TestOps metadata routes with the existing client. Its normal API-token session exchange is the only authentication POST. - Record closed per-row attachment types and flags. Replay the unchanged verifier against the captured responses using historical context as data. Keep the actual process environment unchanged. - Add nine mock and archive-safety tests to the existing Node unit roster. ## Verification - `node --test tests/runner-e2e/chat-feature-testops-retained-verification.node-test.mjs`: 9 passed on Node 26.4.0. - Three Node syntax checks, actionlint, the relevant workflow policy check, and `git diff --check` passed. - The real pinned sanitized export ZIP passes the new safe-export extractor and existing bound reader. The first local attempt used a noncanonical `/var` temporary path and hit the existing canonical-path guard; the canonical `/private/tmp` attempt passed without a source change. - Corrected a review finding in failure attribution. Only actual authentication or protected GET errors now receive `apiFailure`. Historical replay failures retain their original phase and predicate diagnostics. - Initial test run: eight passed and one test expected the generic upload error instead of the existing specific job-run error. Corrected that test expectation. No verifier behavior changed. - Tests exercise separate identities, foreign source and actor rejection, origin and artifact integrity, exact five GETs, all three attachment-row refusals, secret redaction, wrong job/run/source, and real archive traversal/symlink rejection. - The unchanged API and export readers are source-pinned. Existing evaluation and acceptance code is unchanged. - [Normal CI 41220476243](https://github.com/paperclipai/paperclip/actions/runs/41220476243) passed on exact head `b79e85f71567f662d6f3d680ec66ec49a132589a`: 55 successes, three skipped checks, and two neutral reviews. Both required checks passed. Greptile is 5/5 and all review threads are resolved. - The nine new controls are registered in the family unit command. The separate normal PR-CI runner suites do not execute that roster, so their green status is not proof these nine ran on Node 24. The manual workflow explicitly runs them on Node 24 before any protected API read. - The real protected API operation has not run. A separately approved merge and metadata dispatch remain required. No model, app, database, or native build ran locally. ## Risks - The fixed artifacts can expire. Any absent or changed artifact fails before TestOps authentication. - The current protected response snapshot may differ from the original failure-time state. The receipt states that limit and makes no historical cause claim. - This is a fixed diagnostic operation. It does not repair the original workflow or grant authority to run another evaluation. - The workflow uses the existing protected reporting token. It logs no response body, unknown field, credential, or unknown URL. Only closed receipts are retained. ## Model Used OpenAI `gpt-6.1-sol`, through Codex, with reasoning, code editing, shell tools, and source review. The host 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> |
||
|
DottaandPaperclip
|
d38cdf261a |
perf(ui): speed up Tasks and Inbox startup (#18492)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Tasks and Inbox are the main entry points for that work. > - Earlier changes made their data reads and row counts smaller. > - Those changes remain, but shared controls now pull in the whole task-detail bundle. > - A cold open can also read a remembered organization before selecting the one in the URL. > - This pull request separates collection code and selects the addressed organization first. > - It warms task code after collection paint, so the next task open stays fast. ## Linked Issues or Issue Description Refs #17756, #18323, #16849, #17041, #17983, #17960, #18461. **What happened?** Tasks and Inbox still use their optimized index reads and row budgets. However, small shared controls and hooks were placed in the task-detail chunk. A collection had to download that whole chunk. The HTML also preloaded it on every route. The original production profile loaded about 6.5 MB of decoded JavaScript before useful rows, or 1.9 MB transferred with compression. Company bootstrap could select a different remembered organization before Layout selected the URL's organization. This started unnecessary company reads. A retained multi-company browser fixture reproduced two attention-count requests for different companies. **Expected behavior** Show correctly scoped, stable collection rows without waiting for task-detail code. Keep compact reads, bounded rendering, direct task preloading, immediate New Task input, and fast follow-on navigation. **Steps to reproduce** 1. Use a production UI build with Coordinator enabled and two active organizations. 2. Remember one organization, then cold-open Tasks or Mine in the other. 3. Inspect startup script and company requests before the first useful row. The baseline UI revision is `91fdf02e95`. This is a UI change. It has no API, schema, migration, or authorization changes. ## What Changed - Group the static collection dependency trees above the task-detail group. Keep the existing startup and task grouping strategy. - Preload task detail and its static dependency closure early on direct task and agent-chat URLs. Keep collection aliases out of that preload. - Fetch the immutable task chunk at low priority after a collection paints, or on a same-origin task-link hint. After a successful immutable response is fully read, module-preload the cached chunk so it can compile without executing. A failed HTTP probe skips module preload. - Prefer an addressed active organization during bootstrap. Preserve manual selection, archived-company handling, and account-scoped company lists. - Preserve a task reading anchor when a programmatic scroll event reports an unchanged offset during restore reflow. Strengthen the Back journey to wait for a completed collection navigation. - Add production-browser checks for delayed indexes, correct company scope, stable row geometry, failed speculation, and early direct-task preload. Update collection verification notes and the feature map. ## Verification Final source: `e402ab7fc6`; baseline UI: `91fdf02e95`. - All **234 focused unit cases** and **44 production-browser journeys** passed on the final source: 33 task journeys and 11 collection journeys. Coverage includes compact history, saved cards, draft persistence, avatar continuity, failed-probe recovery, delayed data, pagination, deep links, desktop/mobile and reading-position restoration through Back. Desktop and mobile screenshots were inspected. - Workspace typecheck, full production build, map checks, token gates and diff checks passed. The default local `pnpm test:run` still stops before executing tests because the existing chat-test line selector reports `No test found` (1,816 skipped). This is not a passed full local test suite. Focused tests and remote CI remain separate evidence. - Greptile reviewed the exact final commit at **5/5**, with zero open review threads. Final code CI passed. The complete check inventory is **55 successful, 3 skipped, 1 neutral**. The neutral result is the separate Paperclip Review: its agent finished without completing an assessment, so it is not a passing review score. Five alternating baseline/candidate pairs per scenario, Chromium 151.0.7922.34, 1440×900, 4× CPU, 50ms latency, approximately 9Mbps down / 1.5Mbps up. Cold cases use fresh browser contexts with normal cache writes enabled; warm cases use the same context and normal application query cache. Both previews use the server's immutable hashed-asset cache policy and uncached HTML, with real local feature settings and no request interception. All task-owned builds and suites were stopped during timing; other host activity was uncontrolled. Readiness requires useful rows and editable search, or the readable task thread and editable composer, followed by two animation frames. | Journey | Baseline median (range) | PR median (range) | Median paired saving | | --- | --- | --- | --- | | Cold Tasks | 3239 ms (3189–3333) | 2545 ms (2441–2682) | +752 ms | | Cold Inbox | 3207 ms (3203–3268) | 2679 ms (2502–3062) | +541 ms | | Warm Inbox | 717 ms (701–917) | 760 ms (745–768) | -47 ms | | Warm Tasks | 1150 ms (1126–1207) | 1136 ms (1119–1174) | +18 ms | | First task click | 943 ms (932–1008) | 1007 ms (976–1098) | -74 ms | | Back to Tasks | 584 ms (568–617) | 566 ms (559–575) | +10 ms | | Cold direct task control | 3389 ms (3377–3402) | 3415 ms (3395–3896) | -13 ms | All five pairs improved for each cold collection. The signed saving is baseline minus candidate for each pair; it is distinct from the difference of medians. Warm navigation is broadly similar. First task click was 74ms slower by median paired difference. Direct-task median paired difference was 13ms slower, with one slower 3.90s candidate sample retained in the range; this is not a claimed direct-task speedup. Completed script transfers by collection readiness fell from about **1.91MB to 1.14MB** (Tasks), and **1.90MB to 1.14MB** (Inbox). Task code warms after rows paint. At the test bandwidth, the removed 0.77MB critical transfer represents roughly 0.65s of download time; CPU and network overlap, so their savings are not added. Unthrottled loopback API control medians stayed about 17–18ms in both variants. All 30 cold samples had zero browser errors. Rejected experiments remain separate: automatic splitting produced over 200 startup requests; disabling recursive grouping slowed direct tasks; HTTP-only warming improved collection startup but slowed first task opens. The final correction preloads direct dependencies and compiles successfully cached code after paint. Routing disables HTTP caching, so intercepted runs were kept as diagnostic evidence rather than used for warm-cache acceptance; see [Playwright's cache behavior](https://playwright.dev/docs/api/class-page#page-route). Vite preview's default asset `no-cache` policy was also corrected to match the shipped server before final measurement. ## Risks - Bundle placement affects shared consumers. Browser checks cover collections and task navigation; the full PR checks also cover other routes. - Code warming still transfers the task chunk if the user stays on a populated collection. It moves that download past first paint and does not execute it. This is a startup latency improvement, not a claim that all eventual bytes disappear. - A direct click before warming finishes still loads code on demand. Direct task and chat URLs retain early module preload. - Local browser evidence does not establish a deployed staging speedup. This PR does not merge or deploy itself. ## Model Used OpenAI GPT-6 (Codex), with reasoning, tool use and local code execution. The context-window size is not exposed in this environment. ## 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 the relevant tests locally and they pass (234 unit cases and 44 browser journeys; the full-command limitation is 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 code CI gates are green (the separate Paperclip Review is neutral because its agent did not complete an assessment) - [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> |
||
|
|
34f8d25089 |
fix(github): expose review finding categories (#18429)
## Thinking Path > - Paperclip manages AI agents and their work. > - GitHub review bots submit structured assessments through task-bound tools. > - Each repository can set its own finding categories. > - The schema accepts a string, but the server rejects values outside the saved review policy. > - Review instructions did not list the accepted values. The error did not list them either. > - This change gives agents the values before submission and keeps server validation. ## Linked Issues or Issue Description **What happened?** Review submissions failed with an unknown finding category. The agent corrected the assessment before publication. The sampled trace has two rejections followed by a processed review. Historical tool inputs are null. Authorized diagnostic inspection found no saved raw provider trace for the representative run. The exact rejected values remain unknown. **Expected behavior** Agents can find the accepted categories before submission. Unknown values still fail validation. **Steps to reproduce** Set a custom finding category. Start a review. Submit a different category. Before this change, the instructions and error do not list the accepted value. ## What Changed - List configured categories in manual and automatic review instructions. - Return categories from the saved assessment policy in `begin_review`. - Describe the dynamic constraint in the shared discovery schema. Keep custom categories valid. - List accepted values in category errors. Keep authorization and publication checks. - Add contract and database integration tests. Update documentation and access maps. ### Hypothesis and thesis for a fix The missing category guidance can cause avoidable submission retries. A model can also ignore the policy returned by metadata reads; the historical inputs do not establish which occurred. The repair belongs to the GitHub review contract. Use the saved policy categories at review start, explain the constraint, and retain server rejection. A fixed global enum would reject valid custom categories. ## Verification - Baseline `f1d7808`: two new contract assertions failed for missing category guidance and error detail. The same assertions pass after the change. - `pnpm exec vitest run server/src/services/chat-github-review-policy.test.ts server/src/services/chat-github-tools.test.ts`: 41 passed. - `pnpm exec vitest run server/src/__tests__/chat-channels.integration.test.ts -t 'returns snapshot finding categories'`: 1 passed. Real PostgreSQL and service code; simulated GitHub HTTP. Checks policy changes, rejection before publication, accepted custom category persistence, and processed publication. - `node scripts/check-maps.mjs`: passed. Structural evidence only. - Shared TypeScript check passed. Server TypeScript checking was killed by the 8 GiB memory limit, including a separate retry after the other checks stopped. - Full typecheck and build could not complete: `cargo` is absent. Full tests were stopped after concurrent checks exhausted memory. CI must supply the full checks. - Follow-up compiler repair: the unchanged baseline reproduces TS2352 in the category discovery test. Runtime matchers replace the unsafe schema cast. They retain both guidance and custom-category assertions. The repair is commit `0dcb94c676`, with current master merged at `14d7052854`. The final head is `c46b8f35add9255fc71be294d27f5a7865bcb47f`; the remaining code-comment review finding was removed and resolved. - Final-head CI: all gates pass, with no failed or pending checks. Greptile gives this head 5/5 with no outstanding findings; the only inline thread is resolved. The workspace test shard passed on one bounded rerun after an earlier CLI child-server readiness timeout; no code change was made for that rerun. - The review host passed full recursive typecheck and build after the compiler repair and master integration. Fresh database rechecks hit the host PostgreSQL semaphore limit. The full local test command hit a pre-existing chat line-selection failure before assertions. These local limits do not replace the passing final-head CI evidence. - No live model or GitHub publication at the changed revision. No deployment or merge. ## Risks - Instructions add policy text. An agent can still ignore it. - Existing bot catalogs get the schema description when tools are synchronized. Trusted review instructions and `begin_review` also expose categories. - The service test uses simulated GitHub HTTP. It does not prove that deployed agents avoid retries. ## Model Used The original worker used OpenAI `gpt-6.1-sol` through Codex with reasoning, tool use, and code execution. A later Codex review repaired the test compiler error. That review runtime exposes GPT-6 but no exact snapshot 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: devinfoley <139239+devinfoley@users.noreply.github.com> Co-authored-by: Paperclip <noreply@paperclip.ing> Co-authored-by: Dotta <bippadotta@protonmail.com> Co-authored-by: cryppadotta <34892728+cryppadotta@users.noreply.github.com> |
||
|
DottaandPaperclip
|
e09f40888a |
fix(recovery): preserve native task access during resolution (#18489)
## Thinking Path > - Paperclip controls agent execution and task ownership. > - A native run needs its task binding until it reports its result. > - Direct task updates already reject self-status changes that remove this binding. > - Recovery resolution can also change the source task status. > - That route did not use the same protection. > - This change rejects the unsafe status change before the task or recovery receipt changes. > - The run can then use semantic completion while normal recovery permissions remain effective. ## Linked Issues or Issue Description Refs #18411. **What happened?** A running native owner can resolve an older recovery action with `sourceIssueStatus: "in_review"` or `"done"`. The request succeeds and changes its own task status. That can release the execution binding before the run reports. Later task tools fail with `paperclip_runner_tool_binding_not_authorized`. **Expected behavior** Return a conflict before the task status or action receipt changes. Tell the run to report through `paperclip_finish` or `paperclip_block`. Preserve unchanged-status resolution, Board recovery, legacy-agent behavior, and the existing conversation and handoff rules. **Steps to reproduce** 1. Give a running native owner the task execution lock and an older recovery action. 2. Call `POST /api/issues/{id}/recovery-actions/resolve` as that run. Use `outcome: "restored"` and `sourceIssueStatus: "in_review"`. 3. Before this fix, the route returns 200 and changes the status. After this fix, it returns 409 with `native_run_status_reported_by_finish` and makes no task or action mutation. ## What Changed - Reuse the existing native self-status guard inside recovery resolution's source-task transaction. - Let the guard receive the requested status from either route contract. - Add HTTP route regressions for rejected review/completion transitions and continued task editing after rejection. - Cover unchanged-status native resolution, Board recovery and legacy-agent resolution. - Update execution semantics and both recovery maps. ## Verification - Baseline `91fdf02`: the two new rejection cases fail because the real Express route returns 200. Three safety cases pass. - Fixed source: all 198 tests in `server/src/__tests__/issue-agent-mutation-ownership-routes.test.ts` pass. - `node scripts/check-maps.mjs` passes. This is structural evidence. - Greptile and Paperclip Review both give the integrated head 5/5 with no actionable findings. - Full recursive typecheck, full build, and a repeat of all 198 route tests pass at integrated head `0fa104dfb240500acd01303cd446363ea39c6c75`. All current-head CI gates pass (55 successful checks/statuses; no failed or pending checks). - The full local `pnpm test:run` command fails in chat-test line selection with `No test found`, before assertions. This also occurs with the unchanged master chat test file. It is not a passing full local suite. - The route tests use the real route, validation and guard. They mock service and database boundaries. The observed hosted failure establishes the original mechanism. Fixed-source live provider acceptance remains unverified. No deployment or merge is part of this PR. ## Risks - A native owner now receives a conflict for a previously accepted recovery status change. It must use the existing semantic result path. - This prevents a new binding loss. It does not prove that an old remote executor stopped or recover missing provider state. - Board, legacy, conversation and handoff behavior retains existing authority. The regression includes representative allowed paths. ## Model Used OpenAI GPT-6 through Codex, with reasoning, repository tools and code execution. The runtime does not expose a more specific snapshot 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 (198 focused route cases; the full local command has the recorded selection failure) - [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> |
||
|
DottaandPaperclip
|
c38844eafa |
test: observe complete workspace retention cleanup (#18507)
## Thinking Path > - Paperclip manages AI agents and preserves their work. > - Remote workspace retention removes saved objects after their last pin clears. > - The retention worker deletes each object before it commits the final snapshot state. > - The parent-revisit test treats the first object deletion as completion of all work. > - That valid intermediate state can fail the next-deadline assertion. > - This pull request observes complete storage, database, and worker settlement within the existing bound. > - The test still fails if the parent is not revisited after the child releases its pin. ## Linked Issues or Issue Description Refs: #18351 Refs: #18506 **What happened?** [The server6 CI job](https://github.com/paperclipai/paperclip/actions/runs/41263476916/job/124721633674) failed at `remote-workspace-snapshot-retention.test.ts:248`. The test expected a null retention deadline after the first parent object disappeared. The parent could still have its second object and a `delete_requested` row. The worker was not finished. **Expected behavior** The test must wait for both parent objects, both snapshot rows, and the retention worker to finish. It must retain the four-second bound and 40ms observation interval. **Steps to reproduce** 1. Use the parent/unfinished-child fixture in the retention test. 2. Hold the second parent-object deletion after the first object is removed. 3. Read the old assertion while the parent row remains `delete_requested`. 4. The old assertion receives `1793361600000` instead of null. Release the hold and observe the completed state. **Paperclip version or commit** The failed CI checkout was `ab849702a78f6bb619b1527596cd52db402d9189`. This test-only repair starts from actual mainline `b110271c9b2fd6dd5a1beb119e7f0663385df8b9`. The relevant test, retention service, coordinator, storage, test-helper, and configuration blobs match the failed checkout and mainline `756599e4ea069c9338c320fe8846523d3dfb92a0`. **Deployment mode** Built-from-source server tests. The local controls used one disposable PostgreSQL17 instance and local disk storage. They used no deployed company or model. ## What Changed - Observe both parent objects and the exact parent and child database rows in one bounded assertion. - Require deleted rows with cleared deadlines, zero active tracked work, and no coordinator wake before the test ends. - Keep the original four-second timeout and 40ms interval. Production retention and scheduling code stay unchanged. ## Verification - All17 tests in `remote-workspace-snapshot-retention.test.ts` passed on Node24 with current-source aliases, current migrations, and a disposable PostgreSQL17 database. - At immutable head `1e7eeb422be260bfaf752792596f8cdd126fff8d`, the second-deletion hold control also passed all17 file tests. Removing only the parent-revisit signal still failed the affected case within four seconds. The owned database container was stopped and its automatic removal was verified. - Normal checkout command: `pnpm --filter @paperclipai/server exec vitest run src/services/remote-workspace-snapshot-retention.test.ts --reporter=verbose`. - Four bounded controls used the exact failed CI source: unchanged baseline PASS; held second deletion with the old assertion RED; the same held state with complete-settlement observation GREEN; removed parent-revisit signal RED under the original four-second limit. - The held state proved: first parent object absent, second present, child row deleted, parent row `delete_requested`, non-null deadline, and active worker. This proves a reachable observer bug. The exact interleaving in the historical CI job was not recorded and remains unknown. - The same case passed at mainline756 in [the canonical readiness control](https://github.com/paperclipai/paperclip/actions/runs/41260612021/job/124712930107). Source identity and a passing control alone do not establish cause; the held-state control supplies that evidence. - `git diff --check` passed. The net source change is one test file, with no added comments or runtime changes. - Final exact head `1e7eeb422be260bfaf752792596f8cdd126fff8d`: [canonical CI41264965700](https://github.com/paperclipai/paperclip/actions/runs/41264965700) completed successfully. All57 checks are terminal and clean, with no pending or failed checks. This includes the aggregate verify and Snyk gates. The PR is mergeable and clean. No old failed job was retried. - [Greptile reviewed this exact full head at5/5](https://github.com/paperclipai/paperclip/pull/18507#issuecomment-6552310742), with no actionable findings and zero open review threads. Complete thread pagination was checked. - [The current Paperclip assessment is4/5](https://github.com/paperclipai/paperclip/pull/18507#issuecomment-6552292204), with a final full-head footer. Its check conclusion is neutral. It calls the fix correct and identifies one minor, non-blocking concern about a process-wide active-work counter. This score remains4/5; it is not represented as5/5. ## Risks The assertion is stricter because it checks complete deletion and worker settlement. A real missing revisit, retained object, unresolved row, or continuing worker must still fail. The bound is unchanged. No schema, API, production storage, scheduler, or deployed configuration changes are included. No user-facing command or behavior changed, so no product documentation update is needed. The Paperclip review notes that `idleWorkSnapshot().active` is process-wide. The current test configuration uses fork isolation, one worker, and serial tests, so the counter is scoped to this fixture in the present test run. We retain it because deleted rows and an empty wake queue can be observed before the coordinator finishes its active work. If test isolation or fixture concurrency changes, this assertion will need a worker-specific settlement check. This is the review's accepted non-blocking limitation; it is not waived or scored as resolved5/5. ## Model Used OpenAI Codex, GPT-6 family, with reasoning, source inspection, code execution, and tool use. The session does not expose a more specific backend model ID or context-window size; neither is inferred. ## 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>canary/v2026.1101.0-canary.2 |
||
|
DottaandPaperclip
|
b110271c9b |
test: observe GitHub enrollment after worker settlement (#18506)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - GitHub connection events use a worker that must return to quiet after it drains an event. > - The integration test checks that work with real PostgreSQL and signed loopback requests. > - The loopback server can receive an ACK before the worker completes that request. > - The test reads the next deadline too early and can reject valid work in progress. > - This pull request waits for the observed work state without moving the fake clock. > - It keeps the exact work counts, identity checks, deadline bound and quietness checks. ## Linked Issues or Issue Description Refs #18395 and #18351. This is a test-only repair for the event-worker integration fixture. **What happened?** [CI run 41257779002](https://github.com/paperclipai/paperclip/actions/runs/41257779002/job/124704574485) failed at the replacement follow-up deadline assertion. `coordinator.nextWakeAt()` was `null` after the server received the second ACK. The worker can still be waiting for that ACK response at this point. **Expected behavior** Wait for the worker to expose a deadline or finish the exact expected follow-up. Do not drive fake time during an observation. Require the scheduled follow-up to stay within five seconds of the first observation. Keep one empty follow-up, the new identity and no I/O after the worker becomes quiet. **Steps to reproduce** 1. Run the existing enrollment integration test on the original CI merge `a1e25444a95f3698c3761cd1796bd1e11567224c`. 2. Hold only the second loopback ACK response after its server counter increments. 3. The original deadline assertion fails while the worker remains active. **Paperclip version or commit** The causal control used the exact CI merge above. This repair starts from verified master `66e8fab9447aa21350f87ddec79fcd027f3e23ad`. The fixture and relevant worker files match the earlier `b889` inputs. The original failure remains preserved. The controlled race does not prove the unique historical scheduling of that CI attempt. ## What Changed - Use bounded real I/O yields for observations. Assert that the fake clock stays fixed during each observation. - Advance fake time only to actual pending follow-up deadlines. Keep one fixed five-second reference for each drain. - Accept an already finished follow-up only when the exact expected lease count is present. Quietness must still follow. - Hold the replacement ACK on a Promise gate in the existing test. Release it during normal work and in cleanup. This keeps a deterministic regression without a sleep. - Keep the signed request, identity replacement, exact lease counts, 25-minute no-I/O checks and 30-second test budget. ## Verification - Original CI: one assertion failure at line 126; 2,619 other tests passed in that lane. It was not a test timeout or host failure. - Earlier controls: the same test and relevant worker inputs passed in two fresh CI lanes. Both used zero Turbo cache hits. Input identity alone is not the cause proof. - Causal controls on the original CI source: unchanged baseline passed; held ACK failed at the original assertion with active work; a settlement barrier passed all original assertions; a missing second hint still failed. - Final patch with real PostgreSQL 17 and current source migrations: the direct test passes, including its deterministic ACK gate. An extra 750 ms ACK delay also passes. The missing-second-hint negative still fails at the expected observation. - Focused command: `pnpm exec vitest run --project server server/src/__tests__/github-connection-enrollment-work.test.ts`. Local checks used current-source package aliases and one disposable loopback PostgreSQL instance. The test creates a unique database and no user data. - Fresh Greptile review is 5/5 for `519c3fd4a99bf77aea7b085b67c38701358fef95`. The exact head has no open review threads. The current Paperclip assessment is also 5/5 with a final response footer for that full commit. Its check conclusion is neutral. - [Canonical CI 41260540629](https://github.com/paperclipai/paperclip/actions/runs/41260540629) completed successfully for the exact head. All 57 checks are terminal with no failures or pending checks. This includes `ci / verify` and Snyk. The PR is mergeable and clean. - `git diff --check` passes. Full repository typecheck, tests and build are delegated to fresh PR CI. No QA journey, model run, deployment or CI retry was performed for this repair. ## Risks This change affects one integration fixture. Production worker logic, schedules and deadlines do not change. The observer retains a one-second real-time bound and the test retains its 30-second budget. The test still fails on missing work, extra work, a late deadline, a wrong identity or I/O after quiet. The database and HTTP server remain owned by the fixture. ## Model Used OpenAI Codex, GPT-6 family. The session does not expose a more specific backend model ID or fixed context window. Used reasoning, repository tools and 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 - [x] My branch name describes the change and contains no internal Paperclip ticket id or instance-derived details - [x] I have run the focused tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes (test-only behavior; described here) - [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>canary/v2026.1101.0-canary.1 |
||
|
DottaandPaperclip
|
756599e4ea |
fix(workspaces): keep project references without source folders (#18504)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Agents can receive project references in normal task and chat messages. > - Remote workspace preparation checks those references before a native run starts. > - A project with no workspace source has no folder to attach. > - The current check rejects that project as an unsupported read-only folder. > - This pull request keeps the project reference and reports the missing source without attaching files. > - Projects with configured sources keep the current remote access restrictions. ## Linked Issues or Issue Description Refs #18403 and #18351. This is a product prerequisite for the hiring journey. It changes no journey code. **What happened?** A native chat run failed before execution when its message referenced an authorized project with zero configured workspaces. The error was `workspace_validation_failed`: “This remote runtime cannot safely provide read-only project folders.” The project had no repository or checkout. The failed QA run and original grade remain preserved. **Expected behavior** Keep the task workspace when a referenced project has no configured source. Report that no project folder was attached. Continue to reject configured cross-project sources that require an unsupported read-only mount. **Steps to reproduce** 1. Use a native remote task folder with multi-project references enabled. 2. Mention a same-company project that has no configured workspaces in a normal comment. 3. Prepare the run workspace. **Paperclip version or commit** Reproduced on `b889ac39b7fbed558c5f4cfbbc2b73c4ca916810` with the real workspace resolver and PostgreSQL migrations. The staging failure was recorded in workflow [41253428446](https://github.com/paperclipai/paperclip-cloud/actions/runs/41253428446). ## What Changed - Distinguish authorized project references with a known empty workspace list from configured sources. - Return a resolution warning and failure receipt for an absent source. Do not attach a folder or create a workspace. - Preserve project authorization, company filtering, the disabled remote-source gate, and the read-only restriction for configured sources. - Add real PostgreSQL regression cases and document the reference contract. ## Verification - Causal RED on unchanged `b889`: the metadata-only project case failed with the same read-only error. The real repository negative already passed. - GREEN: eight PostgreSQL cases pass. They cover an absent source with the remote-source switch both on and off, a saved existing workspace, a repository, a configured source without a repository URL, the disabled remote gate, mixed references, and a foreign company. - The existing assembled workspace test also passes: a real read-only project source is rejected before reservation or environment acquisition. Total: 9 selected tests pass; 119 unrelated tests were not selected. - Tests use PostgreSQL 17 and the current source migrations. The local Vitest configuration selects current DB, shared, and Runner sources because the borrowed dependency output predates the remote workspace feature. - Current-source TypeScript checking of both changed TypeScript files and their dependency graph passes. It uses the repository's real Express actor declaration. - `node scripts/check-maps.mjs` and `git diff --check` pass. - Canonical reproduction: `pnpm exec vitest run --project server server/src/services/heartbeat/remote-workspace-source.test.ts` with the existing embedded PostgreSQL support or `PAPERCLIP_TEST_POSTGRES_ADMIN_URL` set to a disposable test cluster. - Historical `0f2b` passed all 59 CI checks. A separate non-author manual source audit reported no actionable findings at confidence 4/5; it did not rerun tests or qualify live QA. The official 4/5 review requested a flag-scope test and saved-workspace control. This test-and-documentation update adds both, with production source bytes unchanged. - Current-head canonical [CI 41258109282](https://github.com/paperclipai/paperclip/actions/runs/41258109282) passed: all 59 checks are terminal (55 success, 3 skipped, 1 neutral), with no failures or pending checks. Build and typecheck passed. The eight-case regression file passed in the canonical Linux server 1/12 lane. Its checkout `6f69ccca70d10e78885b133fb54a80940bf175a5` has parents `b889` and this exact head; its tree equals this head. - Fresh Greptile and official Ponytail assessments each give 5/5 for exact `27660027d12ff31427d611019e90dabfb6d8503a`, with zero unresolved review threads. The separate manual audit remains 4/5. - Live QA remains pending on a reviewed app build. This regression proof does not claim that the hiring journey passed. ## Risks A project source could be added after project hydration. This change uses the same run-preparation source snapshot as the existing authorization and source check. It only permits a positively known empty workspace list. Configured sources, unknown shapes, foreign projects, and existing execution authority keep their current checks. No schema, credentials, provider, model, budget, or filesystem permissions change. ## Model Used OpenAI Codex, GPT-6 family. The session does not expose a more specific backend model ID or fixed context window. Used reasoning, repository tools, TypeScript and 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 --------- Co-authored-by: Paperclip <noreply@paperclip.ing> |
||
|
DottaandPaperclip
|
192d83e828 |
Fix chat catalog prerequisite clocks and retained evidence (#18377)
## Thinking Path > - Paperclip evals must prove actual work and retain inspectable evidence. > - Project defaults must reach new work while prior work remains active. > - A hidden HTTP deadline invalidated that prerequisite before the case clock. > - Report rewriting and missing evidence files obscured real outcomes. > - Observer assumptions also rejected valid history, reads and lease bookkeeping. > - This PR repairs those layers and keeps the original feature requirements and clocks. ## Linked Issues or Issue Description Refs #18364 for product repairs. Refs #18170 for protected catalog evidence. The held worker's native fetch had a five-minute header timeout. Successful cases lacked required files. Restart compared stored Markdown to rendered text. Diagnostic redaction changed trusted Playwright program files, causing the existing producer guard to reject failed-report retention. Valid dedicated project reads and automatic environment leases were also misclassified. This PR targets **`codex/chat-catalog-master-integration`**, not master. Its base combines master `51ea624344`, preserved catalog `e2b35be76f` and all product/test changes from #18364, including the native chat response-stage contract repair. The frozen preservation branch remains unchanged. The full catalog stays off master. Shipped subscription aliases retain their 120-second turn limit. ## What Changed - Hold the exact worker with native HTTP, a one-use agent and the original absolute deadline. Keep its capability in an owned 0700 directory / 0600 script outside evidence trees. Put only its path in task/tool inputs. Register the token with the existing secret scanner. Remove file/listener after normal execution and failed setup. - Use the already-selected original native case deadline for prerequisite readiness. Consume only public lifecycle state and fixed recovery diagnostics, preserve manual-repair/terminal failures, and record bounded state transitions and recovery flags. Other lanes retain their authored 45-second setup cap. Nine tests pass actual server diagnostics through the public agent normalizer. - Poll canonical source state with the existing settlement predicate; collect full domain/browser evidence after settlement. - Capture required `final-state.png` and actual `api-state.json`. Keep restart logs inside owned roots and recover abnormal evidence from the actual capture root. - Compare attributed rendered history before/after restart. Independently require exact stored bytes, identity and no duplicate admissions. - Authenticate and scan canonical Playwright program files with the existing validator before preserving their exact bytes. Keep ordinary redaction and every original path, producer, secret, size and transfer guard. Trusted protocol files remain unchanged. - Expose only fixed diagnostics, recognized tool names and bounded counts. Keep raw provider output private. - Accept correlated `list_projects` reads only when the same native call delivered the exact full saved target-project defaults. Cover real OpenCode and server event producers; reject stale, truncated, foreign and unpaired reads. - Version project-defaults as **v4**. Accept natural “newly admitted” future-scope wording; still reject replies missing the future/existing distinction. Exclude only automatic lease acquisition/release with the exact `environment_lease` entity type from settings-change rejection. Retain rejection of actual budget/environment edits, mismatched types and unknown events. Keep active-work, inheritance and reply checks strict. ## Verification **Qualified: [pilot 41256592937](https://github.com/paperclipai/paperclip/actions/runs/41256592937) passes 3/3, with all 35 independent checks passing.** | Case | Checks | Full case duration | Evidence and cleanup | | --- | --- | --- | --- | | Instance restart | 9/9 | 140.1 s | complete | | Project defaults | 11/11 | 204.2 s | complete | | Project work | 15/15 | 204.0 s | complete | All three are original-attempt results from source `c51e079f7f2364305b131ffd549e263ff78c5335`, containing the exact product files from `13c2b61f31d873a475ea0c3c026b19926ff42301`. Actual saved conversation contracts were inspected, as were ordinary native worker contracts where present. Each protected archive includes the actual API state and final PNG; cleanup completes. Original queries, `openrouter/deepseek/deepseek-v4.1-flash-1017`, finite case clocks, concurrency one and zero eval retries are preserved. Earlier passes do not transfer into this qualification. Both current heads pass all checks: product **62** (56 success, one neutral, five skipped), catalog **56** (54 success, one neutral, one skipped). Both have exact-head Greptile **5/5**, zero unresolved review threads and clean mergeability. The three trusted private-evidence protocol files remain byte-identical to the dispatch workflow. Raw evidence stays protected. - **202 focused catalog/evidence tests and catalog typecheck pass.** Nine readiness controls consume actual lifecycle diagnostics through the public agent normalizer, which omits private operation state. Manual repair, terminal states and unknown errors still fail. - Real Node 24 control: old fetch fails at 301,464 ms; native HTTP completes after release at 310,054 ms under the same 340-second control deadline. Actual native OpenCode/Runner also completes a separate source while the held worker stays blocked. Failed-setup regressions verify resource cleanup. - All **45 privacy/report controls pass**. The old failed-report trace rewriting reproduces the producer-guard rejection; authenticated canonical program bytes now survive unchanged. Contaminated assets still fail. No secret, producer, path, transfer, expansion or size guard is weakened. - Oracle **`paperclip.chat-catalog.project-defaults.v4`** retains active-work, inheritance, actual mutation/read and reply requirements. Positive and negative controls cover correlated full saved-default reads, automatic lease bookkeeping and natural “newly admitted” wording. Original failed grades remain retained alongside explicit versioned regrades. - Product validation in #18364 includes 2,044 discovery tests, 112 real-database/HTTP/native-runner/OpenAPI consumer tests, and passing typecheck/build. Task-owned Docker PostgreSQL unblocked local DB checks and was removed afterward. Retained attempt ledger; no earlier result substitutes for the qualified run: | Attempt | Recorded outcome / finding | | --- | --- | | 41114636294 / 41119215255 / 41203017450 | Initial report/setup and unavailable-archive failures. Functional partial success did not qualify the reports. | | [41207155025](https://github.com/paperclipai/paperclip/actions/runs/41207155025) | 0/3. Rendered-Markdown mismatch; other archives unavailable. | | [41211541655](https://github.com/paperclipai/paperclip/actions/runs/41211541655) | Work-only diagnostic: 15/15 with required files and cleanup. | | [41212946325](https://github.com/paperclipai/paperclip/actions/runs/41212946325) | 2/3. Defaults did not settle; transfer failed. | | [41216880087](https://github.com/paperclipai/paperclip/actions/runs/41216880087) | Defaults failed; counters did not identify finish outcomes. | | [41219519639](https://github.com/paperclipai/paperclip/actions/runs/41219519639) | Native work succeeded but grader did not; report assets and expansion limits rejected transfer. | | [41222196525](https://github.com/paperclipai/paperclip/actions/runs/41222196525) | 1/3. Restart exposed fabricated empty-response success; defaults exposed backfill and missing dedicated-read recognition. All archives retained. | | 41226224666 / 41226710271 | Cancelled before paid cells to repair generated outputs, credential custody and native event-shape coverage. | | [41227965230](https://github.com/paperclipai/paperclip/actions/runs/41227965230) | 2/3. Defaults 9/11 correctly rejected existing-task backfill and reply scope. | | [41233723348](https://github.com/paperclipai/paperclip/actions/runs/41233723348) | Restart 9/9. Defaults 10/11: v2 misclassified automatic lease events. Work was cancelled because replacement was dispatched prematurely; no work pass inferred. | | [41236679567](https://github.com/paperclipai/paperclip/actions/runs/41236679567) | 1/3. Defaults 11/11. Restart 4/9: no answer. Work failed and the unchanged ZIP expansion guard rejected full retention; missing outcome remains inconclusive. | | [41240010122](https://github.com/paperclipai/paperclip/actions/runs/41240010122) | Restart 4/9 under authority guidance: its literal native criterion still required the unavailable action. Defaults 10/11: v3 rejected valid “newly admitted” wording; v4 explicitly regrades 11/11 while retaining the original grade and privacy findings. Work 10/15: created/repaired a task without final chat reply. All three archives retained. | | [41243833425](https://github.com/paperclipai/paperclip/actions/runs/41243833425) | Source `6fa0c83674` / product `dddd2d5206`: 2/3. Restart 9/9 and work 15/15; defaults failed prerequisite setup before any canonical query/native run because its 45-second helper cap cut short a normal 60-second probe. | | 41247798414 | Cancelled before paid cells after review caught a private-field assumption in the readiness helper. The repair uses public lifecycle state and fixed recovery diagnostics. | | 41248428190 | Cancelled before paid cells after the unresolved-thread sweep caught a conversation identity edge case. The repair uses canonical identity and carries the user ID through heartbeat. | | [41250502608](https://github.com/paperclipai/paperclip/actions/runs/41250502608) | Source `7b1de24ce1` / product `c0bf249807`: 1/3. Work 15/15. Restart 4/9 despite the repaired contract actually being admitted; it continued discovery after board-only 403 and had no reply. Nearby HTTP 200 reads are not call-correlated to the MCP timeouts. Defaults timed out before settlement (zero finish proposals), then exceeded the unchanged 256 MiB original-evidence bound; detailed domain behavior remains inconclusive. | ## Risks The prerequisite needs native Node in the worker environment. Trusted program bytes remain subject to the existing authentication and scanners. The audit exception rejects unknown events. Oversized ZIPs can still be rejected, as in the retained work attempt; this PR does not weaken that guard or claim an unobserved outcome. A passing pilot qualifies the selected attempts, not full-catalog reliability. ## Model Used OpenAI GPT-6 through Codex, with repository inspection, tool use, code execution and live eval tooling. Exact hosted model revision and context-window size are 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 #` 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 - [x] Final same-source three-cell pilot passes 3/3 with required evidence and cleanup --------- Co-authored-by: Paperclip <noreply@paperclip.ing> |
||
|
DottaandPaperclip
|
66e8fab944 |
fix: complete native chat responses and improve API discovery (#18364)
## Thinking Path > - Paperclip must perform work and report its outcome accurately. > - A chat response and its delegated or operator-owned work are separate stages. > - Native chat contracts instead required literal completion of the user's whole request. > - Agents kept searching or waiting even when a truthful handoff should finish the response. > - OpenCode also has separate acknowledgement backlog and empty-response failure paths. > - This PR aligns chat completion with its current stage and makes OpenCode completion truthful. ## Linked Issues or Issue Description Refs #18098 for retained OpenCode evidence and #18377 for separate eval setup, oracle and reporting repairs. **What happened?** Real pinned OpenCode 1.18.34 reproduced an MCP timeout behind 800 text fragments. Pilot [41222196525](https://github.com/paperclipai/paperclip/actions/runs/41222196525) separately exposed a fabricated successful reply after a reasoning-only response. Later restart attempts timed out without an answer. Their native criterion literally required restarting the instance, although the chat instructions and authority required an operator handoff. General authority guidance alone did not resolve it. **Expected behavior:** A standing chat completes its current authorized response. It performs and verifies permitted actions, verifies delegated handoffs, or explains the specific unavailable action and operator path. The worker's task remains separate. Ordinary execution tasks must still complete their requested work. Empty provider responses must not become invented success. **Version and reproduction:** Actual pinned OpenCode and native Rust Runner, with retained full-stack AWS attempts. Product baseline master `51ea624344`. This PR adds no historical catalog to master. ## What Changed - Build standing-chat completion contracts from the server-owned conversation identity. Scope each current request to the response stage; preserve request text, criterion IDs, order, human-response scope and source revisions. Require verified authorized actions and truthful handoffs. Ordinary task contracts stay strict. Existing admitted inputs retain their saved contract during recovery; new contract content uses the existing revision/hash mechanism. - Guide prospective defaults to their owning resource and preserve existing work unless backfill is requested. Guide unavailable board/operator operations to the supported handoff. These are general chat instructions with no eval identities or expected answers. - Batch adjacent assistant text for at most 50 ms and 4,000 UTF-8 bytes. Preserve text, metadata and order; flush before non-text frames and shutdown. - Give an empty OpenCode response one correction within its existing turn, model, reasoning, context and deadline. Accept actual completed answer parts with incomplete headers. Preserve actual active tool work, bound idle evidence grace and fail truthfully if correction stays empty. - Reject empty OpenCode terminals in Rust without actual prose or an accepted semantic result. Never invent successful replies or evidence. Refresh generated project-discovery contracts. - Preserve accurate public API-discovery descriptions for existing project defaults and managed instance restart operations, including their field scope and authority limits. ## Verification **Qualified: [pilot 41256592937](https://github.com/paperclipai/paperclip/actions/runs/41256592937) passes 3/3, with all 35 independent checks passing.** | Case | Checks | Full case duration | Evidence and cleanup | | --- | --- | --- | --- | | Instance restart | 9/9 | 140.1 s | complete | | Project defaults | 11/11 | 204.2 s | complete | | Project work | 15/15 | 204.0 s | complete | All three are original-attempt results from source `c51e079f7f2364305b131ffd549e263ff78c5335`, containing the exact product files from `13c2b61f31d873a475ea0c3c026b19926ff42301`. Actual saved conversation contracts were inspected, as were ordinary native worker contracts where present. Each protected archive includes the actual API state and final PNG; cleanup completes. Original queries, `openrouter/deepseek/deepseek-v4.1-flash-1017`, finite case clocks, concurrency one and zero eval retries are preserved. Earlier passes do not transfer into this qualification. Both current heads pass all checks: product **62** (56 success, one neutral, five skipped), catalog **56** (54 success, one neutral, one skipped). Both have exact-head Greptile **5/5**, zero unresolved review threads and clean mergeability. The three trusted private-evidence protocol files remain byte-identical to the dispatch workflow. Raw evidence stays protected. - **2,044 API discovery tests pass.** Red controls showed that “restart instance” selected unrelated drain/settings operations and “save project defaults” selected repository replacement. Accurate public summaries now rank the existing operations correctly; route registration preserves descriptions. No endpoint, permission or field behavior changes. - **112 real-database, authenticated HTTP, native Runner and OpenAPI consumer tests pass locally**, using disposable task-owned Docker PostgreSQL. It was removed afterward; shared IPC was untouched. These cover persisted contract selection, admitted native input and the model envelope together. The isolated CI suites also pass. - **77 focused contract/input/context tests pass.** Red controls catch the wrong conversation predicate; canonical identity requires both agent and user IDs, including the actual heartbeat projection. Ordinary tasks, agent-only issues and planner tasks remain strict. Coalesced requests preserve criterion identities, source hashes and human-response scope. - **144 OpenCode driver tests pass**; native core **441 pass, one ignored**; packaging **56 pass, six skipped**. Server typecheck and build pass. - Real pinned OpenCode/native Runner: a delayed 25-second acknowledgement completes within the original 30-second MCP deadline. The text-fragment control changes 817 notifications / 54.1 seconds / timeout to 45 / 19.0 seconds / accepted finish. Separate fast, six-minute and real authenticated API probes complete 32 tool calls plus finish. These do not reproduce every remaining historical MCP timeout, and no common cause is claimed for all such failures. Retained failures remain attributed to their original source. Earlier runs exposed fabricated empty-response success, prospective-default backfill, a literal whole-task chat criterion, fixture readiness and oracle/reporting defects. The immediately preceding [41250502608](https://github.com/paperclipai/paperclip/actions/runs/41250502608), product `c0bf249807` / source `7b1de24ce1`, is **1/3**: work 15/15; restart 4/9 with the repaired contract actually admitted but no answer; defaults timed out before settlement, then exceeded the unchanged 256 MiB original-evidence allowance. Defaults domain behavior on that attempt remains inconclusive. #18377 retains the full attempt ledger, including missing archives and pre-paid cancellations. The fresh 3/3 establishes these selected outcomes; it does not establish full-catalog reliability. ## Risks The shared contract change applies to standing native chats across providers. It changes model instructions and completion scope, while permission, approval and review gates remain enforced by their existing owners. Incorrect model claims remain possible; the product evals independently require actual mutations, native work and delivery. The correction adds at most one provider request within the existing deadline; text batching can delay adjacent text by 50 ms. Three qualified attempts do not establish full-catalog reliability. ## Model Used OpenAI GPT-6 through Codex, with repository inspection, tool use, code execution and live eval tooling. Exact hosted model revision and context-window size are 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 #` 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 - [x] Final same-source three-cell pilot passes 3/3 with required evidence and cleanup --------- Co-authored-by: Paperclip <noreply@paperclip.ing>canary/v2026.1101.0-canary.0 |
||
|
Dotta
|
308d273bb0 | fix: make project defaults and instance restart discoverable | ||
|
DottaandPaperclip
|
b889ac39b7 |
fix: serialize native completion replies and bounded retries (#18502)
## Thinking Path > - Paperclip manages agents and their work. > - A delegated task reports its result to the source conversation through a durable completion outbox. > - Native execution can succeed before the presentation worker saves the accepted final reply. > - The outbox treated that short interval as a missing answer and cleared the target run. > - The presentation worker then rejected the valid answer because its delivery no longer belonged to that run. > - This change retains the existing target for one fixed presentation grace while accepted presentation is pending. > - The existing presentation owner can save the reply without another model run or more budget. ## Linked Issues or Issue Description **What happened?** A completion-delivery retry can run immediately after a native callback succeeds. It clears the callback's delivery target before the presentation worker saves the accepted answer. The reply then fails with `chat_completion_superseded`. A daily run limit correctly blocks the unnecessary replacement run, leaving the conversation active without its saved answer. **Expected behavior** The outbox must retain its current target and attempt while that exact native run has accepted presentation work. The existing presentation worker must save the answer once, acknowledge the delivery, and settle the conversation. Existing actor, session, result, digest, budget and retry checks must remain enforced. **Steps to reproduce** 1. Accept a native DONE result and current-turn final message for a completion callback. 2. Run its due completion-outbox handler after terminal projection but before reply presentation. 3. Run the existing presentation worker. Before this fix it rejects the reply because the outbox has cleared the target. **Paperclip version or commit** Reproduced on `b7f1c3769c00a107ffe4128b48b0ef8cc452cf3f` through the real PostgreSQL outbox and presentation services. The reported failure remains preserved. A new QA run is required after deployment; local tests do not qualify the deployed fix. Related: #17696 schedules accepted-response recovery. #18118 preserves canonical completion receipts. #18364 handles provider output and response-stage contracts. Those changes do not protect the completion-outbox target during this interval. ## What Changed - Defer a new execution attempt when the existing selector finds pending presentation for the same company, conversation, run and agent. - Retain the queued delivery target and attempt with an atomic compare-and-set. Await the existing scoped worker notification before that write in the same transaction. - Serialize retry with reply insertion under the existing child-task-first lock. Recheck locked task Done state and the delivery's stable completion-event version, status, target, attempt and exact lease. A delivered, retargeted or newer-claim winner prevents another wake. Admit the next attempt only after the winning transaction commits and releases its locks. - Limit presentation grace to five existing one-minute delivery leases, measured from the original run `finishedAt`. Clamp both the delivery lease and the native retry deadline to that immutable limit. Repeated sweeps cannot extend it. After expiry, resume the existing retry, exhaustion and daily-limit checks. - Keep publication authority in the existing materializer. The candidate query only defers a retry. Read the freshly reloaded delivery status, target and saved receipt after acquiring its child-task lock. - Fence both hold and retry writes with the exact claimed event version, attempt and lease. A stalled old hold cannot overwrite a newer lease. - Keep completion-event identity separate from mutable task ownership versions. Ordinary Done reassignment preserves its existing completion event; a real reopen retains normal supersession. - Add real PostgreSQL coverage for the observed ordering, outer commit and rollback, one saved reply, conversation settlement, idle SQL, retry boundaries and superseded deliveries. - Document the completion-outbox and presentation contract. ## Verification - Before the fix, the real PostgreSQL regression fails with `chat_completion_superseded` in `existingChatCompletionReply` when the outbox handles the terminal run first. After the fix, the same test passes without a new model wake. - Focused command: `vitest run --project @paperclipai/server server/src/services/native-runtime/agent-chat-response-recovery.test.ts server/src/__tests__/chat-completion-delivery.test.ts` — **95 pass** on `45f53a8c32fcca0bbdacf717286d37e30c1ba308`, using a disposable database on an owned local PostgreSQL 17 container. - Controls preserve retries for failed, cancelled, missing-final, missing-accepted-result, old-generation, foreign-company and foreign-agent targets. Reset or superseded task versions retain supersession. Digest and external-context declines never authorize a reply, never extend the fixed window, and leave it at exact expiry. An external native deadline cannot extend the limit. A valid reply at the last millisecond saves once without another model run. - The first reviewed head `b6f9c484b9567bc60af6d684c04e00058c1048ed` received Greptile 5/5 but an independent 3/5 review correctly found an unbounded hold. Three new real PostgreSQL controls failed on that source. This head adds the fixed bound and makes them pass. The earlier review, controls and CI remain retained; its review score does not qualify the new head. - The bounded predecessor `7300d2a978a5c4b3c13d85899bed3e98ae3c49fa` received an independent 4/5 review that correctly found an expiry-versus-in-flight-reply race. A real PostgreSQL advisory gate reproduced it after reply authority checked the delivery but before insertion. The retry lock and exact claim CAS now make that control pass. Both orderings, outer rollback, delivered/retargeted/superseded winners and competing lease claims pass. The next-wake test can acquire both parent and child locks, proving dispatch occurs after release. The valid 4/5 review and failing control remain retained. - At `39dd86e8f51d937cf382b2757736e0477201d21d`, both reviews found valid remaining defects: the reply helper still used its pre-lock delivery snapshot, and ordinary Done reassignment could stall a delivery by changing the task version. Four real PostgreSQL controls fail on that source and pass here: concurrent retry-first, a newly saved receipt after waiting, and reassignment within grace or at expiry. A real reopen through the issue service still supersedes the old event. - A later coordinator review found the hold CAS still lacked the attempt and lease fences already present on retry. A real PostgreSQL producer barrier reproduced an old hold shortening a newer claim from 120 seconds to 60 seconds. The same exact claim fences now prevent that overwrite. This failed control is retained; no old run was blindly retried. - The real worker saves one exact run-owned reply and acknowledges the delivery only after outer commit. An outer rollback publishes no reply. After settlement, advancing the scheduler clock by 60 seconds causes no further SQL. - Module boundaries, map structure, `git diff --check` and the canonical server build (including TypeScript and Runner vendor preparation) pass. The first local build stopped in the host package-manager setup; the installed pinned pnpm completed the build. Full repository typecheck, test and build gates pass in current-head CI; they have not all been run locally. - Final local checks also caught a test cleanup type error (TS2677) and one 94/95 readiness failure: the worker saves its comment before post-commit conversation settlement finishes. The test-only follow-up removes an invalid cleanup type predicate and waits on the existing bounded worker-quiet check before reading final state. All assertions and the same five-second bound remain. Source `45f53a8c32fcca0bbdacf717286d37e30c1ba308` passes all 95 cases and the canonical server build. The earlier failures are retained. - The final policy cleanup removes only 15 newly added narrative comment lines (seven service, eight test), as required by CONTRIBUTING.md. Both TypeScript AST signatures and `removeComments=true` emitted JavaScript are identical to tested source `45f53a8c32`. This was checked with the TypeScript 6.0.3 API and the installed native TypeScript 7.0.2 compiler. Assertions, timeouts and behavior are unchanged; the 95 tests were not repeated for identical compiled code. - The superseded `45f53a8c32` CI run recorded two 15-second timeouts in attention-retention cases. That failure remains retained. The test and its ten inspected owner or support files are identical on the tested master revisions. The canonical base run passes all 53 attention cases, and the comparison PR passes the same server lane, including those two cases in 9.3 and 9.8 seconds. This does not establish the timeout cause. Current-head CI is a new run after the required comment cleanup, not a retry of the failed source. - The optional Paperclip review bot ended neutral without a completed assessment because its execution paused. That result is not an approval. A separate [independent read-only audit](https://github.com/paperclipai/paperclip/pull/18502#issuecomment-6550216227) reports 5/5 confidence on exact head `9cfd3f74c69cc7a2da7b4ee5d1fca80f15d4af0c`, with no findings. Its manual assessment does not replace or relabel the incomplete bot result. - Current head `9cfd3f74c69cc7a2da7b4ee5d1fca80f15d4af0c` has a fresh Greptile 5/5 review with no open threads. [CI run 41249007430](https://github.com/paperclipai/paperclip/actions/runs/41249007430) passes all primary jobs and both required gates (`ci / verify` and `ci / e2e`). All 59 PR contexts are terminal, with no failed contexts. The actual CI checkout is `18db9e2750d6a799f26b5b8ba31bcccbea254227`, which combines master `681f7b77262a9df6c1b4aad55291bf960061e491` with this exact head. Fresh QA verification on the deployed application remains required before claiming the live failure is fixed. ## Risks A candidate can retain a delivery whose stricter presentation checks still prevent publication, but only within the fixed five-minute window. The existing worker keeps publication authority. If a valid reply remains unavailable at expiry, the normal callback retry may require another admitted model run; the original result and completed child work remain preserved. Existing retry, exhaustion, model-run and spending limits still apply. Failed or cancelled execution and genuinely missing accepted output retain the existing retry path. ## Model Used OpenAI GPT-6 through Codex, with repository inspection, tool use and code execution. The runtime does not expose the exact hosted revision 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>canary/v2026.1031.0-canary.18 |
||
|
Dotta
|
cdfd701d89 | fix: use canonical conversation identity for native contracts | ||
|
DottaandPaperclip
|
681f7b7726 |
perf(server): retire aggregate maintenance polling (#18425)
Reviewed and verified at
canary/v2026.1031.0-canary.17
|
||
|
Dotta
|
0d8ac2855e | fix(chat): scope native completion contracts to the response stage | ||
|
DottaandPaperclip
|
9ca9af544a |
Drive remaining live run and resource recovery from durable work (#18419)
Reviewed and verified at
canary/v2026.1031.0-canary.16
|
||
|
DottaandPaperclip
|
eb12cfc96b |
perf(github): activate connection work from events and deadlines (#18395)
Reviewed and verified at
|
||
|
Dotta
|
3e89f82783 | fix(chat): hand off operations outside the current identity | ||
|
Dotta
|
6ba01c52b3 | test(runner): exercise correction deadlines with virtual time | ||
|
DottaandPaperclip
|
5e4cc01073 |
feat(workspaces): keep remote files off the webserver disk (#18403)
Keep native sandbox workspace files independent of the webserver disk. Use encrypted saved versions and Postgres ownership so a replacement server can recover files and continue work. Enable the path by default for new eligible workspaces when durable storage is configured. Preserve explicit opt-outs and existing host workspaces. Include safe snapshot repair, Files access, import, migration and retention. Bound saved-Files resources, preserve imported Git recovery, reject unfinished Git operations, publish immutable concurrent captures, and verify permanent object-version deletion. Validated 471 targeted local controls, build/typechecks, all 40 CI partitions, fresh Greptile 5/5 and security review. The PR records original failed attempts and scoped retries. Co-Authored-By: Paperclip <noreply@paperclip.ing>canary/v2026.1031.0-canary.15 |
||
|
DottaandPaperclip
|
75fa37f30f |
fix(server): expire native backups from retention deadlines (#18380)
Reviewed and verified at
|
||
|
Dotta
|
55be2a8bc8 | fix(runner): settle corrected answers from completed parts and actual tool activity | ||
|
DottaandPaperclip
|
453f5fbb78 |
perf(external-objects): schedule scoped refresh obligations (#18384)
Reviewed and verified at
canary/v2026.1031.0-canary.14
|
||
|
DottaandPaperclip
|
5bc029bf6c |
test: wait for native work and accounting before Stop redirect (#18463)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Product E2E journeys test the browser, native work and saved outcomes on a selected deployment. > - The Stop journey must interrupt real work before it sends a new direction. > - Its old boundary accepted a turn-start event before any command or usage report. > - A stopped attempt with incomplete usage correctly blocks later work under the existing budget policy. > - This pull request waits for a live native command and bound usage, then checks actual accounting settlement before it sends the redirect. > - The provider reports first usage after its initial command completes, so the fixture finishes a harmless readiness call before the long interruptible hold. > - The benefit is a stronger test of Stop and continuation with the same three-run limit. ## Linked Issues or Issue Description Refs: #18098, #18345, #18354. Related accounting work: #18035. Integrated upstream prerequisites: #18387, #18465 and #18409. Duplicate selector PR #18472 was closed as superseded by actual merged #18409. **What happened?** A QA attempt clicked Stop soon after a native turn started. The owned event stream had no command or usage report. Its cancelled run retained incomplete accounting. The saved redirect received a skipped wake, and no third run started. The budget guard must keep that missing usage visible. **Expected behavior** The journey must first prove that actual native work is active and that its current turn has reported complete usage. The fixture must finish a harmless readiness command before it starts a separate bounded hold. After Stop and physical cleanup, it must verify a complete accounting receipt and saved settlement before sending the redirect. **Steps to reproduce** The fresh staging attempt failed before Stop because its first 90-second command ended before the provider emitted its first complete usage. Its only later eligible commands lasted less than one second. The controlled browser sequence reproduces that ordering without accepting it. Completed readiness usage followed by a separate still-active hold qualifies under the unchanged oracle. Startup alone, command alone, usage alone, partial usage, incomplete accounting and a missing settlement timestamp still hold the next action. ## What Changed - Request a harmless `printf` readiness command first. Wait for its result, then request `sleep 90` exactly once in a separate shell call before file edits. This gives the provider an opportunity to emit current-turn usage before the interruptible hold. - Reuse the strict native-command observer. Require a bound current-turn complete usage report with valid nonzero token counts. Reject stale, partial, foreign and semantic-tool evidence. - Wait for the stopped run's complete accounting receipt, `accountingReceiptReady: true`, `costAccountingPending: false` and a valid saved settlement time before redirecting. A receipt already ready before Stop is intentional only when the exact cancelled run is actually accounted with no pending debt; no source/sequence advance or callback clock order is assumed. - Exclude canonical raw, partial and complete `item.completed` usage only when the API row and PRP envelope match the authoritative owned run, company and agent, and their source event ID, instance ID and positive sequence agree. Missing context keeps the strict work check. A self-consistent accounting source can change after Stop. Keep the original snapshot and reject real tools, turns, messages and other item events, including productive output from a new source. - Retain private boundary and accounting proof. Source and sequence observations do not impose unsupported controller or clock order. Include the new oracle in the suite definition hash. Keep existing success-check IDs, exact three runs, model, budget, deadline, Stop proof and final outcome checks. - Reuse one binding validator in the active-work boundary and post-Stop filter. Pre-Stop keeps its current Runner/current-turn restriction. Add focused evidence tests and a standalone Chromium/control-API sequence check. The optional pre-Stop hook leaves existing classic callers unchanged; they do not call the accounting filter. ## Verification - Current source is `47578138ad357fc11b8b346bfada36d6aa5ab484`, one narrow fixture commit on reviewed `b9668ccfba61c59eb496ba18269aee06e2921d8f`. Only the prompt constant, two ordering controls, the existing Chromium sequence and the journey documentation changed. All oracle functions, run counts, model, budget, deadline and check IDs are byte-identical to b966. There is no new upstream merge or app change. - Fresh staging failure is preserved in the [sanitized report](https://pages.paperclip.ing/evals-qa-smoke-41227920892-1/). It measured harness `5d3f143d0aed4dc7cf472b7891ea3d4d95ab843b` against app `b7f1c3769c00a107ffe4128b48b0ef8cc452cf3f`; the reviewed Stop helpers and its flow were unchanged in that harness. The original grade has `stop_boundary_already_settled`, two succeeded native runs, and complete cleanup. Authentication, setup, collection and cleanup passed; the journey failed. This attempt remains failed. - Causal timing evidence: the first shell command started at event108 and completed at111 about 90 seconds later. The first structured complete current-turn usage arrived at113, after the hold ended. Replaying the exact retained event metadata never qualifies that hold. The only later eligible command intervals lasted about 0.90 and 0.18 seconds. The emitter publishes positive usage when the provider sends its assistant-message usage update. The fixture now completes a harmless first call before requesting the long hold; no usage is supplied or invented by setup. - On current source, all 137 focused tests and journey contracts passed. Use `node node_modules/vitest/vitest.mjs run --config tests/runner-e2e/vitest.config.ts product-journeys/stop-redirect-evidence.test.ts product-journeys/agent-evidence.test.ts product-journeys/active-followup.test.ts native-event-envelope.test.ts product-journeys/contracts.test.ts --maxWorkers 1`. New controls reject every prefix when first usage arrives only after the hold completes. They accept the same-turn sequence of completed readiness, genuine complete usage and a separate active hold, and reject it after that hold completes. - Manual Chromium qualification passed: `node --import ./cli/node_modules/tsx/dist/loader.mjs tests/runner-e2e/stop-redirect-boundary.browser.mjs`. The browser does not click Stop for the failed ordering or completed readiness alone. It clicks once for a separate active hold with bound usage. It sends one redirect only after actual settlement. Foreign usage, productive output and missing accounting still fail. This isolated control uses controlled responses and does not call a model or the deployed app. - Strict scoped TypeScript, maps, diff and secret checks passed. The official credential-free Node24 QA runtime-import preflight passed without network access. Full local E2E typing previously encountered old borrowed workspace declarations; clean CI typecheck passed on b966. A full local repository build/test was not repeated for this four-file fixture change. Fresh canonical CI must supply the clean full gates. - Historical binding proof remains valid: the exact old d499 filter failed 17 of the final61 Stop tests. The shared validator checks authoritative run/company/agent plus consistent row/envelope source identity. Bound raw/partial/complete accounting is allowed after Stop, including a consistent source change; productive output remains strict. Pre-Stop still requires its current Runner and turn, complete nonzero usage and a running command. - Historical b966 CI `41224054720` passed all59 terminal checks, with exact-head Greptile and Paperclip Review each5/5 and no open findings. These are not approvals of current475. Current475 has fresh exact-head Greptile5/5 and Paperclip Review5/5, with zero unresolved findings and no omitted thread pages. Canonical CI [`41229185200`](https://github.com/paperclipai/paperclip/actions/runs/41229185200) passed on exact475. All59 checks are terminal:55 success,3 expected skips and1 optional neutral review, with zero failed or pending checks. No old job was retried. Fresh normal staging run41229965802 passed. Its [sanitized report](https://pages.paperclip.ing/evals-qa-smoke-41229965802-1/) agrees with the independently read final original grade:11 checks, all5 phases, exactly3 native runs and cleanup passed. The final manifest is complete, with cleanup complete and no pending evidence. Smoke, report publication and Slack outcome jobs all succeeded. - The live pass measured harness `6be20bef6d0c5077adc1ae99562ad84e756ab167` against app `b7f1c3769c00a107ffe4128b48b0ef8cc452cf3f`. Its Stop oracle, flow, shared binding validator, unit cases and Chromium control are byte-identical to reviewed475. The shared agent-flow file adds only the hiring import and early hiring dispatch; all remaining bytes match. This records the actual harness and app identities separately. - Canonical run readback confirmed the first run succeeded, the second was cancelled and the third succeeded. The cancelled run had a ready native-producer accounting receipt, no pending accounting and a saved settlement time. After verified file copyback, Core queued final destruction and followed its normal five-minute cleanup backoff. The exact lease naturally reached expired/cleanup-success with a bound destroyed receipt before the third run started. No operator cleanup, source change, timeout extension or workflow retry was used. The earlier failed staging attempt remains failed. - Current475 is ready for the normal reviewed merge and QA admission process. Source stays frozen; merge and configuration changes belong to that process. ## Risks - Some provider profiles may not expose a valid current-turn usage report while the command runs. The stronger boundary fails closed in that case. - Accounting may remain incomplete after Stop. The redirect stays unsent within the original deadline. This change does not waive missing debt, treat it as zero, raise a budget or change provider behavior. - This repairs the test boundary. It does not prove or repair a general provider shutdown-usage issue. A separate local late-usage ordering control is not attributed to this QA incident. ## Model Used - OpenAI Codex, GPT-6 family. The exact deployment variant and context-window size are not exposed in this session. Used code reasoning, repository tools, test execution and browser automation. ## 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> |
||
|
Dotta
|
ac75e8fe39 | fix(chat): preserve existing work when changing future defaults | ||
|
DottaandPaperclip
|
c32ecf6657 |
fix(server): schedule only outstanding merge confirmations (#18382)
Reviewed and verified at
|
||
|
Nicky Leach
|
6484f96c28 | fix(workspaces): inspect storage only when needed (#18410) | ||
|
DottaandPaperclip
|
9a3f395737 |
fix(server): settle routine history without idle scans (#18399)
Reviewed and verified at
canary/v2026.1031.0-canary.13
|
||
|
Dotta
|
3937215a05 |
Qualify explicit native written references in the basic question journey (#18459)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Basic task journeys must prove that people can answer an agent and receive saved work. > - The question journey asks for a choice, then a typed reference, then a saved note. > - A native choice card can offer a real written-answer control as well as predefined options. > - The old observer rejected that form before a person could type the reference. > - This pull request accepts that explicit native control and keeps exact answer and work evidence. > - The separate text-only and named-tool qualifications keep their stricter contracts. ## Linked Issues or Issue Description Refs #18346 and #18456. The integration inherits the reviewed fixture prerequisite #18387 and Agent Chat recovery #18465 through their actual merged parent. The producer fix in #18456 is separate from this basic journey contract correction. Public duplicate PR and issue searches found no other correction for this form equivalence. **What happened?** The basic question journey requires the reference card to have `answerMode: text`. It rejects a native single-select card even when its canonical form explicitly enables a written answer. This makes the observer test a widget shape in addition to the requested behavior. **Expected behavior** The person must make a real choice and type the complete reference. A native card with an explicit written-answer control can satisfy the second step. Selecting a predefined option cannot satisfy it. The agent must resume the bound work and save both actual answers in its own task document. The document must survive reload. **Steps to reproduce** Use the focused recording and real QuestionForm browser test in this PR. Keep the reference as a single-select native question with one predefined option and `customAnswer.enabled: true`. The previous grader rejects `question-resume.forms`. The previous browser helper rejects `single_select` before it uses the written-answer control. **Paperclip version / commit** Original source base `f2db909393de124caf6a5380a52d4e0e1d621661`. Current integration `a417fd827347ed6f7998774a9de6aee70b76d867` merges exactly `12b0cc630471ee151ec5bc8e41a34f09236d6723` into reviewed question source `faec4ce6d775a34d303d200987e4e84d6dc26caf`. The scheduled failure on `82d15946c5844d3ae41c6ade62fa224053f20a9a` remains a failure. Its saved card had no canonical written-answer capability and would still fail this change. The retained normalized events do not prove whether the original provider flag was omitted or false. **Deployment mode** Attached native task journeys, including QA Cloud. This PR changes the observer and grader. It does not change the app, provider prompt, model, budget, timeouts, cleanup, or runtime deployment. ## What Changed - Accept one provider-bound single-select reference card with explicit canonical custom answers and valid, distinct options. Keep pure-text reference support. - Make the browser select the actual written-answer control and fill the real rich input. Opt in only for the basic question journey. - Keep the exact no-option typed response, immutable card, board actor, native request and continuation, saved agent document, reload, model, environment, checkpoints, and existing one-to-three-run checks. - Reject missing or disabled custom capability, inferred legacy capability, malformed or extra questions, foreign native identities, changed cards, substituted options, and wrong or missing reference text. - Add a real production QuestionForm and TaskChatRichInput Chromium probe. It first selects a predefined option, then proves the helper clears it and submits only the exact custom text. - Document the contract correction and preserve the historical failed attempt and its separate cleanup outcome. Keep necessary-question and named-tool documentation qualification unchanged. ## Verification - Before the observer change, the focused grader regression failed only the new valid-written-reference case; the other 108 cases passed. - Before the browser helper change, real Chromium passed the text control and four rejection controls but failed the valid native written-answer control at the old pure-text assertion. - Initial focused Runner E2E unit selection at `af07f4de`: **256 passed across 9 files**. This includes question/resumption, clarification evidence/flow/collection/deadline, necessary question, documentation, contracts, and report catalog tests. - Initial real Chromium selection at `af07f4de`: **12 passed**. Six new production QuestionForm/control checks and all six existing document-readiness checks pass. No form or rich-input component is mocked. - Historical test-only head `faec4ce6d775a34d303d200987e4e84d6dc26caf` adds a direct canonical-absent legacy `freeText` grader negative and renders the actual legacy presentation in the browser. All four production helper blobs remain identical to the reviewed head. A temporary legacy inference mutant makes the new negative fail. After restoring production source, **179 focused unit tests and all 6 real QuestionForm checks pass**. The six document-readiness specs and their helper are unchanged. - Original [CI run 41205748589](https://github.com/paperclipai/paperclip/actions/runs/41205748589) failed the staged-listen Runner handoff test; that failure is preserved and was not retried. Its Runner source closure was identical to an earlier passing canonical lane. The symptom also recurred on other heads and is tracked independently; no infrastructure attribution or live pass is claimed. - Current integration `a417fd827347ed6f7998774a9de6aee70b76d867` preserves **all 12 question files byte-for-byte** from faec. Every other tracked path is identical to exact merged parent `12b0cc630471ee151ec5bc8e41a34f09236d6723`, which contains actual prerequisite `baf651c8d0122efc93f3ef03e891919eeb666215`. There is no manual conflict resolution or additional production edit. - On that integrated head: **257 focused Vitest tests across 9 files, 206 required Node source/reporter tests, and 12 real Chromium tests pass**. Map/module-boundary checks and catalog discovery pass. The actual `--preview` command passes with a synthetic question target and an intentionally unavailable adapter; it performs no credential read, adapter loading, provisioning or model call. It retains maxRuns3, a 600000ms campaign bound and zero retries. - Exact-head [CI run 41210806950](https://github.com/paperclipai/paperclip/actions/runs/41210806950) passes, including official typecheck, build, all eight browser shards, both Runner Vitest lanes and aggregate verify. All 59 current PR contexts are terminal with no failures. Fresh [Greptile summary](https://github.com/paperclipai/paperclip/pull/18459#issuecomment-6543878408) explicitly reviews `a417fd827347ed6f7998774a9de6aee70b76d867` at 5/5 with no findings; its exact-head check succeeds and all threads are resolved. No fresh deployed journey pass is claimed yet. - Catalog discovery lists the existing question journey. Map checks and module-boundary checks pass. Diff checks pass. - `tsc -p tests/runner-e2e/tsconfig.json` was attempted. The fresh integrated checkout still lacks compiled Runner, plugin SDK, and adapter declarations. The attempted local check reports 344 diagnostic lines from dependency/type closure and zero diagnostics in all 12 question-owned paths. Official exact-head CI establishes the full typecheck/build gates. Full repository build, typecheck, and test suites were not repeated locally for this observer change. - The first broad local report selection lacked the CLI loader and then hit sandbox IPC restrictions. Locked dependency installation and an authorized local IPC run resolved those setup errors; the full selected 256 tests pass. The initial localhost probe also needed the normal local-listen permission. - No fresh QA pass is claimed. The coordinator must qualify the final reviewed runtime and harness together. Prior failed results remain preserved. ## Risks The basic journey accepts one additional form with explicit native provenance. It still rejects the original closed card without a written-answer capability. The separate producer default fix must reach the selected remote provider pack before it can affect a deployed run. Local browser and grader tests do not prove that artifact served or that a model completed the live journey. Existing cleanup and physical-stop obligations remain separate. ## Model Used OpenAI Codex, GPT-6 family, with reasoning, source inspection, browser testing, and code execution. The host does not expose the exact underlying model version 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 available 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 related work above - [x] I have described the issue in-PR following the bug template - [x] I have not referenced internal/instance-local Paperclip issues or links - [x] My branch name describes the change and contains no internal ticket or instance-derived details - [x] I have run the selected tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation - [x] I have considered and documented risks - [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 |
||
|
DottaandPaperclip
|
91fdf02e95 |
perf(tasks): defer completed transcripts on compact opens (#18461)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Task conversations must show the saved instruction and recent replies quickly. > - Compact task reads already limit the saved message window. > - The conversation still waits for completed run transcripts even when a saved reply contains the result. > - Removing that wait alone can insert activity after the reader starts reading. > - This pull request defers that output until the reader opens history and keeps required output in the initial reveal. > - The benefit is a useful conversation sooner, with a stable first layout and a working composer. ## Linked Issues or Issue Description Refs #17593, #18092, #18312, #18358, #18422. **What happened?** Opening a task with a long saved investigation can wait for native events or legacy logs before showing its existing replies. A slow transcript request can hold the conversation behind the loading screen until the 15-second fallback. **Expected behavior** Show the bounded saved conversation without reading completed output that already has a usable saved reply. Keep that output available through the existing history controls. **Steps to reproduce** 1. Create a task with many completed runs and long saved agent replies. 2. Delay its run events and logs. 3. Open the compact task directly. Before this change, the transcript read blocks the saved replies and composer. ## What Changed - Defer completed successful run transcripts on compact task opens when a usable saved reply is present. Load that output through the existing history controls. - Keep required output on the initial path: live and failed runs, yielded runs, pending interactions, placeholder/progress-only replies, explicit anchors, chats, and full history. Retain output already selected during this visit when its run completes. - Keep saved plans visible when their owning run output is deferred or still loading. - Keep the same composer and draft through history expansion. Offer **More run output** when a single saved run has deferred output. - Position same-task comment links after their target arrives. Wait only for a missing target, with a 15-second limit. Unrelated stalled metadata cannot block an available target after the reveal deadline. - Preserve the reading anchor when a queued scroll event arrives during a properties-pane width transition. - Add a 200-reply/200-run PostgreSQL browser fixture with a saved plan, covering both task views on desktop and mobile. Correct stale browser setup for current routes and the shared mobile scroller. - Remove a random company-prefix collision in the isolated privacy test fixture that failed during CI. ## Verification - Final revision `fba3f2fcbb`: 386 focused thread/scroller unit cases and all 32 production-browser journeys passed. The browser cases cover both task views on desktop/mobile, saved plans, delayed transcripts, history, comment links, Back, drafts, retry, avatar continuity, media and pane resizing. The 25 privacy fixture cases also passed against PostgreSQL. - Workspace typecheck and full build passed with Node 24. A broader local UI run passed 10,220 cases across 849 files before the final navigation/plan review fixes; the changed components were then rechecked in the 386-case focused run. - Five alternating pairs compared master `12b0cc6304` with this revision, using 200 long saved replies and 200 completed native runs. Both production UI builds used the same loopback API/database, Chromium 151.0.7922.34, 1440 × 900, fresh browser contexts, service workers blocked, no CPU throttle, and a controlled 3-second events/log delay. Builds and other suites were stopped during measurement. No page errors occurred. | Useful conversation + editable composer | Baseline median (range) | Fixed median (range) | | --- | --- | --- | | Cold direct open | 3.570s (3.538–3.616s) | 1.056s (1.022–1.141s) | | Task-list navigation | 3.373s (3.342–3.394s) | 0.899s (0.885–0.915s) | | Back | 22ms (22–25ms) | 25ms (25–28ms) | - Median paired savings: 2.519s cold, 2.470s from the list. Completed-output requests: two per baseline sample, zero per fixed sample. Recent-comments API control medians were 13.6ms for both, with ranges 12.1–18.7ms and 11.6–50.8ms. No Back improvement is claimed; an earlier revision's run also included one 1-second Back outlier, retained in the local evidence. - The full local default test command did **not** pass: it stopped on the existing chat test line-selection error. After restoring embedded PostgreSQL library symlinks omitted by an install with scripts disabled, Node 24 passed 672 cases in the first two chat shards, then failed to select the third. This is separate from the passing focused suites and remote CI. - Earlier CI failures were addressed: the privacy prefix collision is fixed; the delegated task-completion reply case passed locally after merging master's reply-recovery fix (#18465). Browser fixture corrections retain the original behavior assertions. Red tests reproduced the loading, navigation, resize and saved-plan regressions before their fixes. - Map, module-boundary, token-gate and added-line token checks passed. The full-tree token scan still finds pre-existing active-username fixtures; this PR adds zero forbidden token lines. - Greptile: **5/5**, final revision, all four review threads resolved. The final-revision PR CI workflow passed, including build, typecheck, all server/chat/workspace test shards, all eight E2E shards, and every Runner lane. GitHub reports 55 successful checks and three skipped checks. The separate **Paperclip Review** check remains queued with **Waiting for the assigned agent**; it has not produced an assessment and has not been bypassed. The branch has no merge conflicts. ## Risks - The default compact task omits completed transcript details when the saved result is sufficient. The reader must open history to inspect them. - Required structural data remains inside the initial reveal boundary. Slow server responses for that data remain visible latency. - Link positioning falls back after 15 seconds when its target never arrives. The conversation remains readable. - The measurements use a controlled local delay. They do not establish a deployed staging speedup. - No API, schema, authorization, or migration changes. ## Model Used OpenAI Codex, GPT-6, 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 - [ ] I have run tests locally and they pass (focused suites pass; the full default wrapper has the existing selection error 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 - [ ] All Paperclip CI gates are green (code CI passed; the separate assigned-agent Paperclip Review remains queued) - [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> |
||
|
Dotta
|
17e22d0bcc | build(runner): refresh project discovery semantic contracts | ||
|
Dotta
|
f947f5d67a | fix(runner): clarify project defaults do not backfill active work | ||
|
Dotta
|
b0d0674fe7 | fix(runner): recover empty OpenCode replies without fabricating success | ||
|
DottaandPaperclip
|
dce383c733 |
perf(server): stop recurring attention retention scans (#18388)
Reviewed and verified at
canary/v2026.1031.0-canary.12
|
||
|
DottaandPaperclip
|
b6b57079b9 |
fix: evaluate status cards on committed changes and deadlines (#18401)
Reviewed and verified at
canary/v2026.1031.0-canary.11
|
||
|
Dotta
|
1e423103b3 | Record refreshed master recovery baseline for catalog pilot | ||
|
Dotta
|
e7522f48bb |
Merge commit '51ea62434413df7a0485e8b0c5f6700671b698a5' into codex/chat-catalog-master-integration
* commit '51ea62434413df7a0485e8b0c5f6700671b698a5': fix(server): reject a native run's own task status change (#18411) Make saved execution receipt recovery react to durable work (#18409) Drive issue recovery from committed changes and exact deadlines (#18402) |
||
|
DottaandPaperclip
|
e40de0c9e8 |
perf(ci): refresh test weights and narrow graph dependencies (#18464)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Contributors need fast, reliable CI feedback. > - Diff-aware CI selects test files and partitions them into runner jobs. > - Stale cost estimates put several slow suites on the same runner. Broad imports also couple small tests to large modules. > - This PR refreshes test costs, runs independent graph discovery in parallel, and narrows three dependency boundaries. > - Selection safeguards and CODEOWNER-only AWS routing stay in force. ## Linked Issues or Issue Description Refs #18398. This follows the audit of [PR #18427's verification run](https://github.com/paperclipai/paperclip/actions/runs/41169978784). The audit found 403 selected general-server files without timing weights. Three suites measured at 232s, 122s and 116s shared one shard because each received a roughly 2s fallback weight. Local graph discovery across four projects took 43.7s serially and 21.7s with two workers. A reverse-order repeat took 37.7s serially and 22.6s with two workers. Both comparisons produced identical selection results. This is a local measurement, not a claim about production CI speed. ## What Changed - Refresh general-server costs for 1,285 of 1,309 current files and serialized costs for all 163 files from two successful protected-master Cloud readiness runs. Retain prior costs for unmeasured files, including dedicated chat and native-runner suites. - Record passing server file costs, including collection, setup and environment preparation. A weekly hosted workflow validates trusted source runs and opens a timing-only PR. It tests shard coverage, pauses while a refresh PR is open, and never merges automatically. - Discover at most two project graphs concurrently in isolated, heap-limited Node processes. Cache repeated inventory reads within one selection. Worker errors and timeouts retain the full task plan. - Fetch only the merge commit and parents for the selector. A real shallow-clone regression proves the complete diff still includes more than 300 files, renames and deletions. - Share the identical log viewer and run details between both AgentDetail pages. The viewer test imports the shared implementation and runs each of its 13 distinct cases once. - Extract the pure API error class while preserving the API client's public exports. Import delivery work signals through their existing narrow DB export. ## Verification - PASS: `pnpm -r typecheck` and `pnpm build`. - PASS: 284 CI selection, timing, shallow-diff, sharding and routing tests. - PASS: 74 further cache, routing and graph-proof tests. - PASS: 94 targeted UI assertions across the viewer, run invocation, API errors/client and progress helpers. All 76 assertions in the 12 server suites made independently selectable also pass. - PASS: `pnpm check:token-gates`, module-boundary checks, YAML parsing and `git diff --check`. - PASS: [Full PR CI and required `ci / verify` aggregate](https://github.com/paperclipai/paperclip/actions/runs/41206683345), including all 12 general-server shards, all nine serialized shards, workspace groups, chat suites, Runner lanes and build checks. - PASS: [GitHub hosted real-workspace selector and cache/routing qualification](https://github.com/paperclipai/paperclip/actions/runs/41206683205). - PASS: Both Greptile and Paperclip reviews are 5/5 with no actionable findings or inline comments on `7dc65c51b8`. The PR has no merge conflicts. - PASS: The production refresh parser accepted all 107 structured file-cost records from a real successful server CI job, including setup/loading cost. - Local `pnpm test:run` completed all six chat shards successfully. The duplicate monolithic local run was stopped after the complete remote suite passed; it was not completed locally. The focused log-viewer command improved from 18.84s (26 duplicate variant cases) to 8.49s (all 13 distinct behavior cases). The dependency audit confirms 12 server files and the API-error test no longer need conservative always-run retention. Repartitioning the audited PR with the new costs changes the slowest shard from 810s to 591s of summed historical test execution (1,124 of 1,142 files measured). This is a workload simulation, not measured new CI wall time. ## Risks - Two graph processes use more peak memory. Each has a 2 GiB heap limit and a three-minute timeout. Any failure preserves the original task plan. Real workspace qualification runs in GitHub CI. - Historical logs contain test/hook costs only. New reporter records add loading/setup costs after merge. The manifests identify measurement coverage and source runs. These weights change placement, never permission to skip tests. - Runner load adds timing noise. The refresh uses medians and requires at least 90% measurement coverage in each lane. - GitHub may require approval to run checks on a PR created with `GITHUB_TOKEN`. Refreshes remain reviewable and do not merge themselves. - The shared viewer is a mechanical extraction. Both pages retain their existing exports, and the behavior assertions remain covered. AWS runner authorization is unchanged. ## Model Used OpenAI Codex, GPT-6, with repository inspection, reasoning, code execution and GitHub tools. The exact deployment ID and context-window size were 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 tests locally and they pass (focused checks and six chat shards; complete suite passed in CI as detailed 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> |
||
|
DottaandPaperclip
|
bacbab2bc1 |
test(testops): retain safe verification failure predicates (#18470)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Chat feature evaluations send their original results to TestOps. > - A final reader checks the stored case, run, source, status, and attachments. > - Two real chat cases passed, but the final reader returned a generic mismatch. > - Later server responses passed the same reader, so they do not explain the earlier failure. > - This change records the exact failed predicate without recording field values. > - The next failure can be diagnosed in its actual caller context. ## Linked Issues or Issue Description Refs #18462, #18394, and #18322. The current base includes the independent browser-fixture fix in #18409. **What happened?** The post-upload reader returned `chat_testops_upload_mismatch`. Its closed artifact had no phase or field detail. The original evaluation and TestOps case passed. Later responses from the same five endpoints passed the unchanged reader. Those later interactive responses do not prove what the protected API caller received at failure time. An earlier CI run also exposed a flat-text browser-fixture mismatch. Current master independently fixes that assertion in #18409. This branch adopts that complete upstream test; the final PR diff contains only the two diagnostic files. **Expected behavior** Keep every report-reader acceptance check. Report a fixed phase, failed predicate names, match booleans, and field type names. Do not record response values, credentials, unknown URLs, or traces. **Steps to reproduce** 1. Supply a missing case ID, a conflicting job ID, or a wrong export field to the reader. 2. Project the failure with `failureProjection`. 3. The parent returns only the generic error code. This change also identifies the failed check and distinguishes a missing field from a value of the wrong type. **Paperclip version or commit** The latest retained failure used |
||
|
DottaandPaperclip
|
b7f1c3769c |
feat(native): record verified provider-pack session admission (#18469)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Remote native sessions can use an installed pack or a retained earlier pack. > - The current app revision does not identify the pack that a session admitted. > - Core already verifies the selected pack and binds its launch artifacts. > - Normal run records lack a closed receipt for that admitted selection. > - This pull request records that receipt after session admission and attachment. > - Operators can compare the actual Core admission identity with their expected pack. ## Linked Issues or Issue Description Refs #17749 for protected provider-pack selection and #18343 for retained controller recovery. The related OpenCode producer repair is #18456. Those changes do not publish the closed run receipt added here. A search found no existing `nativeProviderPackAdmission` PR. **What happened?** Core can admit a compatible installed provider pack with a source revision that differs from the current app. A retained session can also keep its original pack. The normal run record does not show the admitted pack digest, proxy digest, and source revision. An expected launch path or current app revision cannot fill that evidence gap. **Expected behavior** The run has a closed, server-owned receipt after actual session admission and attachment. The first receipt remains immutable history. A separate current-proof state verifies only the actual current admission epoch. Retained recovery needs positive original provenance, and missing legacy evidence stays unknown. **Steps to reproduce** 1. Select and verify an installed pack whose digest differs from the current app pack. 2. Open the native session successfully. 3. Inspect the normal run profile for a bound admission receipt. **Paperclip version or commit** The original selected base was `8bcc5a01edd5d87540401521e6f30d3df43e1848`. The branch now includes exact actual prerequisite parent `3ee4c5b6138da82d0b1f51aa010540561bf43c22`; the PR diff against it is the five receipt, executor, test and documentation files. **Deployment mode** Core remote native execution with provider-pack selection and retained session attachment. ## What Changed - Retain the verified selected pack identity only after a backend returns the matching admitted session. - Publish `runnerProfileJson.nativeProviderPackAdmission` from the session callback after attachment succeeds. - Mark the receipt with `stage: session_admitted` and `authority: core_verified_selection`. - Store bound company, agent, run, session, runner, logical `runnerEnvironmentLeaseId`, pack digest, OpenCode proxy digest, and source revision fields. Retained attachments identify `inheritedFromRunId`. The logical Runner scope is distinct from the physical environment lease. - Lock the current coordinator claim before the profile write. Check owner, attempt, generation, controller boot, lease expiry, phase, and absence of an accepted result. Merge JSON fields without replacing other profile data. - Restore retained metadata only from the exact prior closed Core receipt, positively verified prior state and matching retained binding. Capture same-run prior provenance from the locked claim row before invalidation; cross-run recovery reads only the exact source. Legacy absence remains unknown while recovery continues. - Preserve the first receipt as history. Atomically mark a new opaque admission epoch pending in the existing mandatory claim transaction. Publish current verification only for the matching claim and epoch after actual attachment; changed or unsupported admission becomes unknown without blocking authorized execution. - Preserve synchronous session publication and quarantine ordering before the best-effort receipt write. Fresh replacement selects its newly verified pack; a captured retained-process flag cannot relabel the replacement with an older pack. Preserve the existing private pack authority even when a local dirty revision is ineligible for the closed receipt. - Add focused admission, recovery, identity, PostgreSQL ownership, lock, and JSON-preservation tests. Document the receipt contract. ## Verification - RED: the actual managed-backend path verified and admitted an installed pack but had no closed metadata. Failed open published no evidence. Additional controls fail when retained recovery invents missing history or dirty-revision eligibility gates the existing private authority. - RED calibration: removing mandatory pending invalidation fails the actual pre-admission control; reusing the retained flag for fresh replacement falsely restores the old pack. Removing the epoch fence fails the real PostgreSQL stale-callback case; reintroducing a receipt-specific execution error fails both authorized Runner/logical-lease replacement controls. - GREEN: 49 focused Core tests, all 33 admission identity/transaction tests (20 pure and 13 PostgreSQL), and all 23 existing private-authority tests passed on exact integrated head `165a9096cf3e0d34d2db1e6b52960ad4ac0ad132`. No all-skipped run is counted. - Final test-only head `11ac34baf00552a5d322b5f5f1a8e491859cacec` adds the positive executor-callback SQL-writer wiring control, exact session/identity/claim/epoch assertions, profile preservation, and null-session zero-write check. Removing only the production writer call makes it fail on a missing receipt. All70 focused identity/Core tests pass with this closure; production code, the13 PostgreSQL tests and the23 private-authority tests are byte-identical to the qualified165 source (106 distinct passing controls across the combined proof). - The PostgreSQL tests used an owned loopback-only cached Postgres 17 container with the supported `PAPERCLIP_TEST_POSTGRES_ADMIN_URL` helper and aliases to current DB/shared source. The container was removed afterward. They cover immutable history, JSON preservation, stale claims and epochs, expired leases, terminal runs, changed/unsupported replacements, and a failed write retaining pending proof. - Historical pushed `f7f12fe` CI executed all 24 earlier admission tests, including nine PostgreSQL cases, in server5. Its warm-quarantine failure exposed the new hook-ordering issue fixed here. The same run also encountered a six-character UUID fixture-prefix collision and the independently repaired cached issue-link browser selector; those historical failures are not claimed as green. - Node 24.22.2 and Vitest 5.0.3. The credential-free normal QA runtime import preflight passes with networking disabled. Map contracts, token gates and `git diff --check` pass. - Local TypeScript validation still sees stale borrowed Runner/shared declaration mismatches outside the changed ranges. No diagnostics remain in the new helper or changed test/hook ranges; clean current-head CI is required. - Repository-wide typecheck, full tests and build were not completed locally. A full local executor attempt reached this host's unsupported native PostgreSQL listener and was interrupted; its result is not counted. Canonical run `41214512239`, attempt 2, is successful on exact final head `11ac34baf00552a5d322b5f5f1a8e491859cacec`. All 59 checks are terminal: 55 success, 3 skipped and 1 neutral; none failed or pending. The actual Linux suites execute all 33 admission tests (including 13 PostgreSQL cases) and all 1,022 executor tests successfully. The [original five failures and bounded unchanged-source controls](https://github.com/paperclipai/paperclip/pull/18469#issuecomment-6545593767) remain preserved. One failed-only request was made; GitHub expanded the reusable workflow call and dependents. No second request was made. Fresh exact-head Greptile review is 5/5, all review threads are resolved, and the PR is mergeable. - These local results claim no live provider admission, rollout or recovery. ## Risks - The authority is Core's verified selection plus successful session admission. It is not an independent provider statement or continuous runtime measurement. - Receipt fields cannot grant execution or replace the existing provider, process, lease, workspace, or recovery checks. - Legacy missing receipts and unavailable database reads remain unknown. The current release cannot label an older admitted pack. - The closed receipt requires a clean, 40-character hexadecimal source revision. Verified local `-dirty` packs still execute and retain their existing private recovery authority; their closed admission proof stays `unknown`, so it cannot certify an immutable source revision. - The first receipt is historical. Consumers must require current state `verified`; pending or unknown cannot certify the current pack. Changed Runner scope or pack evidence does not add execution policy. - A receipt save failure leaves the atomically invalidated state pending. The existing artifact authority, cancellation, workspace and process checks remain responsible for execution safety. - The receipt adds no manifest, environment, credential, or filesystem path to the run record. ## Model Used - OpenAI GPT-6 through Codex, with reasoning, code editing, and tool use. The session does not expose an exact model build ID or context-window capacity. ## 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 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>canary/v2026.1031.0-canary.10 |
||
|
Devin FoleyandPaperclip
|
51ea624344 |
fix(server): reject a native run's own task status change (#18411)
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - Native runs (the Paperclip Runner) report their outcome with
`paperclip_finish` or `paperclip_block`, and the status arbiter then
sets the task status
> - Every runner tool, `paperclip_finish` included, requires that the
task's `executionRunId` is the current run
> - When the run sends its own status change through `PATCH
/api/issues/{id}` (for example with `call_api`), the issue update clears
`executionRunId` and `checkoutRunId`, so the run removes its own
execution lock
> - After that, every tool call fails, the run cannot report, same-run
recovery spends its retries, and the task goes to `blocked`
> - This pull request rejects that status change with a 409 before it
commits, and the message tells the run to report with `paperclip_finish`
or `paperclip_block`
> - The benefit is that the run corrects itself in the same turn, and
the task does not get stuck
## Linked Issues or Issue Description
No public issue exists. This is the bug description.
**What happened?**
An agent on the Paperclip Runner followed task instructions that said
"set the task to `in_review` and keep it assigned to yourself" while it
waited on a scheduled monitor. It sent `call_api` with `PATCH
/api/issues/{id}` and `{ "status": "in_review" }`. The server applied
the change and cleared the execution lock. Then:
1. `get_task`, `set_task_monitor`, and `paperclip_finish` all failed the
run binding check. The provider saw only "Paperclip semantic operation
failed", or the monitor-wait message from `paperclip_finish`.
2. The `call_api` receipt for the PATCH showed `status: null, outcome:
unknown`, because the authority check after the response failed when the
lock was gone.
3. The session ended with no semantic result. Exact-checkpoint resume
ran out of retries (`native_session_retry_exhausted`), and recovery
would not replay the run because the mutation outcome was unknown. The
task went to `blocked`.
This occurred on each check where the agent tried the status change.
Runs that did not try it completed normally.
**Expected behavior**
A native run that holds the task's execution lock cannot remove its own
lock. It gets a clear, actionable error and reports its outcome through
`paperclip_finish` or `paperclip_block`.
**Steps to reproduce**
1. Assign a task to an agent that uses the Paperclip Runner (native
runtime).
2. While the run executes, have it send `call_api` with `PATCH
/api/issues/{id}` and `{ "status": "in_review" }` for its own task.
3. Look at the next tool calls of the run. They fail, `paperclip_finish`
cannot complete, and recovery blocks the task.
**Paperclip version or commit**
`master` at
|
||
|
DottaandPaperclip
|
3ee4c5b613 |
Make saved execution receipt recovery react to durable work (#18409)
Reviewed and verified at
canary/v2026.1031.0-canary.9
|
||
|
Dotta
|
76a3af8a43 | docs(evals): record the refreshed chat recovery baseline | ||
|
Dotta
|
f1602d654e |
Merge remote-tracking branch 'origin/master' into codex/chat-catalog-master-integration
* origin/master: fix(chat): recover committed native Agent Chat replies (#18465) perf(heartbeat): drive retry promotion from durable deadlines (#18387) fix(ui): make the Coordinator composer footer opaque (#18455) fix(testops): verify unexpanded result job runs (#18462) |
||
|
DottaandPaperclip
|
42fd568cf8 |
Drive issue recovery from committed changes and exact deadlines (#18402)
Reviewed and verified at
|
||
|
Dotta
|
12b0cc6304 |
fix(chat): recover committed native Agent Chat replies (#18465)
## Thinking Path > - Paperclip manages agent work and keeps its results available to users. > - Agent Chat needs a visible reply before the conversation can wait for more input. > - Native result commitment and reply presentation are separate obligations. > - A completed conversation can retain a valid final message without a saved reply. > - Existing reply recovery excludes these completed conversation turns. > - This pull request repairs that persistence gap through the existing recovery worker. ## Linked Issues or Issue Description **What happened?** A native Agent Chat turn completed successfully and saved its workspace. Its accepted result and final assistant message were durable, but no reply appeared in the conversation. The chat stayed active. The deployment hiring journey in #18351 then reached its deadline while waiting for the first reply. **Expected behavior** Recover the exact accepted turn's final reply once, then use normal conversation settlement. Keep later user input active. Do not run the agent again or replace a selected or deleted reply. **Steps to reproduce** 1. Create an active conversation with one admitted native turn. 2. Commit an accepted DONE result and a final-channel assistant message for that exact run, session and turn. 3. Finish workspace finalization before live heartbeat reply presentation. Leave no saved reply. 4. Run normal response recovery. Before this fix it finds no conversation candidate and leaves the chat active. **Paperclip version or commit** Reproduced with actual services at frozen master `8bcc5a01edd5d87540401521e6f30d3df43e1848`. The production predicate also exists in the deployed revision that exposed the failure. Related work: #18351. The broader receipt and recovery work in #18118 and #18409 does not touch these source files. ## What Changed - Discover unanswered committed Agent Chat turns through the existing native response presentation worker. - Recheck company, agent, responsible user, current execution and conversation generation under the normal transaction locks. - Require the accepted result digest and an exact final message from the same native run, session and accepted turn. Keep external publication authority separate. - Commit the reply and presentation receipt before normal conversation settlement. Later input stays active. Recovery does not repeat agent work. - Preserve skip decisions and selected or deleted replies. Add database controls for foreign identity, stale sessions, partial commitment, duplicates and ordinary internal task behavior. - Document the recovery boundary in the developer guide. ## Verification - Real-service RED/GREEN uses all 418 canonical migrations in isolated PGlite and the exact new regression fixture. Before the fix: zero candidates, no reply, recovery returns false and the chat stays active. After the fix: one candidate, one exact reply, recovery returns true and the chat waits. A repeat returns false and creates no second reply. - All 20 canonical regression cases pass locally through a fixture-only PGlite bridge. Only local database provisioning and `createDb` selection differ from the normal embedded PostgreSQL fixture. Production services, SQL queries, transactions, protocol validation and assertions remain real. Linux embedded PostgreSQL CI is required before merge. - Focused current-source no-emit typing passes for both production modules and the new regression file, with normal Express augmentation and source aliases. The broader local presentation/ownership and conversation-settlement PGlite bridge run was interrupted before it reported results; it is not counted as a pass. No repository-wide local build is claimed from borrowed workspace dependencies. Clean-build CI and canonical PostgreSQL consumer tests remain required. - `git diff --check` and the scoped source/test secret scan pass. No captured run data, credentials or private target identities are committed. - No live journey retry, deployment, configuration change or test-plan admission was performed. The original failed journey remains failed; this production fix still needs deployed qualification through the unchanged journey. - Current-head [canonical Linux CI](https://github.com/paperclipai/paperclip/actions/runs/41207018345) passed all 57 terminal checks with no failure or pending check. The new 20-case database regression actually passed in [server shard 8](https://github.com/paperclipai/paperclip/actions/runs/41207018345/job/124555006090), at 13:59:25 UTC (2,643 ms). Typecheck, build and the existing consumer lanes passed. Fresh [Greptile review](https://github.com/paperclipai/paperclip/pull/18465#issuecomment-6544208002) on exact `f32a432e10b377ae6277d6f728995d5fedcf9d7b` is 5/5; the review-thread inventory is empty with no pagination. No live deployment or journey pass is claimed by these checks. ## Risks Eligible active conversations can gain a previously missing reply. Recovery accepts only the exact saved final message under its accepted result and current conversation identity. It does not run a model, repeat hiring or change a prior test grade. Reply persistence precedes waiting-state settlement; existing conversation recovery retains ownership of a later interrupted settlement. No schema or public API changes are included. ## Model Used OpenAI Codex, GPT-6 family. The session does not expose a more specific backend model ID or fixed context window. Used reasoning, repository tools, TypeScript and real database 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 |
||
|
DottaandPaperclip
|
baf651c8d0 |
perf(heartbeat): drive retry promotion from durable deadlines (#18387)
Reviewed and verified at
|
||
|
DottaandPaperclip
|
f46c8071fb |
fix(ui): make the Coordinator composer footer opaque (#18455)
## Thinking Path
> - Paperclip helps people manage AI agents and their work.
> - The Coordinator workspace shows a task transcript and a message
composer.
> - The composer footer can cover part of the transcript on desktop.
> - The footer has a transparent background, so text shows through
accessory panels and gaps.
> - This pull request gives the footer a solid background from the
active theme.
> - Users can read the composer controls without text from the
transcript behind them.
## Linked Issues or Issue Description
**What happened?**
The Coordinator composer footer lets transcript text show through the
area above the message field.
**Expected behavior**
The full footer has a solid surface in light and dark themes.
**Steps to reproduce**
1. Enable the Coordinator workspace.
2. Open a task with enough messages to scroll.
3. Show a composer accessory, such as queued messages or a pending
question.
4. Scroll the transcript behind the footer on desktop.
**Paperclip version or commit**
Base commit:
|
||
|
DottaandPaperclip
|
47f81e850b |
fix(testops): verify unexpanded result job runs (#18462)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Chat feature evaluations send their results to TestOps. > - The final reader checks the result against its current GitHub run. > - A real result omitted the nested job object and had no presentation links. > - The reader rejected that passed result after a successful upload. > - This change joins the embedded job-run ID to the separate job-run response. > - The reader keeps the source, case, status, and attachment checks. ## Linked Issues or Issue Description Refs #18394 and #18322. **What happened?** The native chat test and cleanup passed. TestOps stored a passed result and three attachments. The final read-only step failed with `chat_testops_upload_mismatch`. The result response had a job-run ID, launch ID, and current GitHub run URL. It did not include the nested job object. Its presentation links were empty. **Expected behavior** The reader must accept that response when the separate job-run response proves the exact job, launch, and current GitHub run. It must reject conflicting identity or source data. **Steps to reproduce** 1. Return a valid result with an unexpanded job run and empty links. 2. Return the matching full job run from the existing job-run endpoint. 3. Call `verifyChatUpload`. The old reader rejects the result. **Paperclip version or commit** Observed oncanary/v2026.1031.0-canary.8 |