mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-29 12:59:13 +01:00
DottaandPaperclip
1aa3a21ac4
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Chat users receive a notice when accepted input waits for execution. > - Periodic publication maintenance scans those receipts even when nothing changes. > - Admission, cancellation, source removal, and authority changes already define when a notice can change. > - Existing action, receipt, publication, and provider-link rows preserve that work across restart. > - This PR gives ordinary inbound queue notices a commit-driven owner using the shared scheduler. > - Settled notices stop querying while unfinished batches and contention retain recovery. ## Linked Issues or Issue Description This ordinary inbound-notice slice follows merged #17895 and now targets master. Related extraction: #16187. Failed-run retry notice projection remains a separate following slice (#17649); its two shared receipt-writer notifications move forward here because ordinary input can attach to those receipt owners. **What happened?** Publication maintenance scans accepted input to project queued, not-started, and removed-source notices. Empty scans keep touching the database. Its cursor also loses PostgreSQL timestamp precision, so a full page of suppressed candidates can repeat. **Expected behavior** Committed changes should wake the notice owner. The worker should reconstruct saved candidates after restart and stop scanning when settled. It must retain a wake across drain, failed selection, contention, and a commit during reconciliation. **Steps to reproduce** Start an otherwise idle instance with chat history. Observe the ordinary inbound notice scan in periodic publication maintenance. For the cursor defect, retain a full page of denied candidates with microsecond timestamps before an eligible notice. ## What Changed - Register an ordinary inbound-notice owner with the existing delivery coordinator and scheduler. Bound each pass to 100 candidates, preserve precise continuation cursors and earlier-page contention, and reset scoped discovery after committed changes. - Notify from real admission, failed action settlement, source removal, task/tree/agent/budget/company cancellation, signed Slack Stop, and deferred queue cancellation/failure. Preserve pending-status and absent-run guards so concurrent promotion wins. - Cover task privacy/grants, project membership/visibility and current authority through their persistence owners. Keep receipt ownership, current authorization, provider policy and closed notice text in the existing chat service. - Cover shared retry-owned receipts through the existing durable authorization callback and a notifying transaction around conditional denial retirement. These small producer fixes already exist in #17649; no per-caller recovery wrappers are added. Retain its matching budget-refusal fixture correction. - Remove only ordinary notices from periodic maintenance. Preserve the separate failed-retry pass at 50 and keep the explicit combined manual API. - Preserve timestamp precision, fairness, restart, drain/resume, rollback and quiet behavior. Add two real producer transaction regressions and update the audit. Add no storage or scheduler infrastructure. ## Verification Current publication head: `41f7e2c9e351efa6cfa5ef699c0e69df7425012b`, after normal merge of master `d30c8f42b842327ee1c18c41a2efd6ba623da5cd` (#17960). The integration preserves both complete added/removed patches, all non-overlap upstream blobs/deletions, and task-before-comment lock order before the notice source lock/intent. Full binary identity is not claimed for the overlapping files. The final follow-up changes acceptance documentation only. - **202 distinct combined-master cases pass on `1720b685d9d5fb9cb4377223fdaa0a9fae2aff9c`:** 148 across the real PostgreSQL receipt concurrency and two comment route suites, plus 54 notice/FIFO/producer cases (1,598 skipped). Current server typechecking/build and module boundaries pass; all 21 owned source/test hashes are unchanged after the documentation follow-up. - The earlier CI voice-reconstruction quiet assertion failed once after 105 fetches. All 15 voice lifecycle cases pass unchanged locally; no assertions or production code were relaxed and no CI retry was used. The original failure remains in the audit, and current-head full CI must pass. - **Earlier restoration repair (head `28c898e93504318056d7a6ab2bbc5eafa6b639ae`): 126 distinct selected cases pass (54 ordinary-notice/producer cases with 1,598 skipped, plus 72 separate Retry/budget cases with 1,580 skipped).** Fix the blocking [restored input stays silent](https://github.com/paperclipai/paperclip/pull/17647#discussion_r4404007181) finding. A failed Retry's earlier receipt wake can drain before `releaseUnadmittedTaskRetryInput` restores ordinary receipts in a later transaction. The restoring transaction now awaits the existing company/task notification before its guarded write. - The new causal regression builds a durable unadmitted failed-Retry owner, uses the real input-claim owner, consumes an earlier committed wake, and proves two minutes of worker quiet with no notice. The real release transaction then restores the ordinary row, emits intent/commit settlement, and reactivates the coordinator to create its queued notice. It fails before the hook with the restored row visible but no intent. This is real PostgreSQL producer/consumer boundary proof with seeded Retry-refusal state; the separate compatibility selection exercises existing supported GitHub Retry flows. It is not a claimed complete public-request failure walkthrough. - **Earlier discovery-repair acceptance (revision `4a713edfa06b9d3dcc0e81e900006fbdb973b2d9`): 267 distinct selected cases pass:** 142 complete cancellation/app/admission consumer cases, 53 ordinary-notice cases, and 72 separate retry/budget cases. The selections skip 1,598 and 1,579 other cases. The final master integration leaves this complete patch byte-identical; the new master paths concern provider workspace/session reuse and E2E reports. - Fix the valid P2 [discovery rereads denied history](https://github.com/paperclipai/paperclip/pull/17647#discussion_r4386889825), despite the previous head's 5/5 summary. Task hints now need zero discovery SQL; company/startup discovery returns DISTINCT company/task identifiers with one query. Bounded projection still checks current authority and candidates. Three real database regressions with 250 denied actions and a second company fail before the fix and pass afterward, checking actual query and returned-row counts. Broad SQL may still inspect candidate history; no constant database-cost claim is made. - Both new producer tests fail before their fixes and pass afterward. One observes intent before a callback-owned receipt write and settlement after outer commit. The other uses real heartbeat admission/recovery and a PostgreSQL row lock to prove exact company/task intent exists before conditional denial retirement; the outside connection still sees deferred status until commit. - The reader can select a retry-owned deferred receipt for later ordinary input, and receipt resolution follows its owner. Disabling coalescing on the retry's own request does not exclude that later path. The two transferred notifications close the actual writer gaps; current authorization remains enforced by existing consumers. This is producer/transaction proof, not a claimed complete mixed-ingress walkthrough. - The moved callback exposes an old test mock that authorized against the root DB. The existing #17649 fixture correction authorizes and writes in one transaction; original requester, comment, skipped-status and budget assertions remain. Both the initial invalid lock fixture and the corrected missing-intent failures are preserved in the audit. - Updated shared dependency build, server typechecking/build and module-boundary checks pass. Because master also changed agent configuration, upgrade admission and runner contracts, the same 264 cases were rerun on combined revision `875a6ee8d211cba7f0499aace7fcea7f6880c4a2` with new isolated databases. Repeated checks are not 528 different tests. Tests under `src/__tests__` were executed separately from production typechecking. - [Current acceptance and preserved failure ledger](https://github.com/paperclipai/paperclip/blob/41f7e2c9e351efa6cfa5ef699c0e69df7425012b/doc/plans/2026-10-25-inbound-wake-notices-sleep-audit.md#shared-retry-receipt-producer-gap-resolved-october-28) records current scope, exact commands, actual fixtures, source hashes and remaining limits. Earlier baseline/full-suite counts retain their original revision scope. - Server typechecking/build and module boundaries pass after the current restoration hook. The earlier combined-master compilation at `661f2f6da8c6c22cdcc25edccd57502ee6fa8c4b` retains its original scope. - **Current exact-head CI is green and Greptile is 5/5.** The old head's review error is not a pass. No full local repository matrix, live provider tenant, external alarm, physical suspension or savings claim is made. Following notice duties retain their separate owners. ### Final merge verification - Reviewed PR head remains `41f7e2c9e351efa6cfa5ef699c0e69df7425012b`: exact-head Greptile 5/5, all 55 check contexts clean, no unresolved review threads, and no Ponytail verdict present. - One Runner shard initially failed descriptor suspension and a five-second OpenCode fixture deadline. The same package passed on the preceding head. The first local attempt was invalid because required Rust fixture binaries were absent; after building them, all eight relevant cases pass with unchanged timeouts. One failed-job retry then passes 1,956 cases (four skipped). No source/assertion/time-budget change or second CI retry was used. - Current master `0e365768f032a472807b5720da8431f92d2f07ce` adds manual monitor and task-drain fixes. Its routes file overlaps but merges cleanly. Local normal-merge validation commit `2720b79c769cb609ad4d392512331e67b910ff3f` preserves both complete added/removed patches and every unrelated master blob/deletion. This is not raw binary identity for overlapping files. - **156 distinct cases pass on that exact combined tree:** all 102 monitor/drain/policy cases plus 54 notice cases (1,598 skipped). Types and build pass. An initial competing-cold-check run failed setup hooks (38 passed, two hook failures, 62 skipped); the same four suites pass alone with all original hook/test timeouts and assertions intact. The failed run remains recorded, not counted as full acceptance. - This merge proof changes no published source or review head. The protected pinned-head merge guard recomputes both patches and requires the exact tested master/tree/head, successful validation receipts and unchanged live merge gates. Previous acceptance counts retain their original revision scope; no local full-repository or deployed-sleep claim is made. Master then advanced to `75754ffb7af2b7ee7dce1e915a145faa82a8fb2a` with #17976 (five UI/e2e/map files only). The entire `server` and `packages` trees are identical to the validated composition. Both complete patches and all newly landed UI files are preserved in the final clean merge tree; the protected gate checks these exact source-root hashes and changed paths. No additional UI-test result is claimed. ## Risks - Missing a supported producer could delay a notice until restart. Tests cover real service and API producers; durable rows remain authoritative. Notifications cover supported application writers in the owning process, not arbitrary external SQL writers. - Privacy invalidation can affect descendants, so those changes use company scope. Ordinary admission and publication changes use task scope. Settled history causes no recurring scan. - Provider output remains subject to current authority and receipt ownership. A cancelled request without a visible queued notice does not acquire a new failure message. - This establishes worker quiet, not host or database suspension. Cloud alarms and conservative sleep gates do not change. Other maintenance remains in place. - Remaining stacked slices still need their own integration, review, and CI. This slice merges only under the user’s explicit campaign authorization after current-head gates pass. ## Model Used OpenAI Codex, GPT-6 family, with repository inspection and code execution tools. The exact serving variant 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 --------- Co-authored-by: Paperclip <noreply@paperclip.ing>
Feature modules
A feature module under server/src/modules/<name>/ uses three layers. Each
layer has one rule: which layer below it, it may import from.
adapters → application → domain
domain/holds pure business rules. A domain file takes plain data in and returns plain data out. A domain file must not importdrizzle-orm,@paperclipai/db, a service underserver/src/services/, a route underserver/src/routes/, or a Node.js I/O module (node:child_process,node:fs,node:net). A domain function must not read the system clock; the caller passesnowas an explicitDatevalue.application/holds use cases and the ports they need. A use case takes its ports as constructor arguments and calls domain functions for policy decisions. An application file must not importdrizzle-orm, a SQL client, a concrete adapter, or the server's HTTP error helpers. The outer service or route translates application errors into transport responses.adapters/holds the concrete implementations of the ports: Postgres queries, transactions, and process control. An adapter file may importdrizzle-orm,@paperclipai/db, and Node.js I/O modules.
A module exposes its commands through index.ts, which composes the adapters
and the use cases behind a factory function. Code outside the module imports
that entry point, never a file inside domain/, application/, or
adapters/ directly.
pnpm check:module-boundaries enforces these rules for production source
files. It also checks access to the company deletion entry point below.
Company deletion
The existing company deletion service owns the database transaction and deletion
order. A module can expose a separate company-deletion.ts entry point.
Only services/company-deletion.ts can import this entry point.
The entry point exports an implementation of CompanyDeletionParticipant from
lib/company-deletion.ts. Do not export it through the normal module index.
The service must complete external cleanup before it starts the transaction. It must lock the company before it deletes dependent records or calls a module. Creation commands must use the same company lock. The service removes dependent records before it calls the module. The module checks its deletion conditions and deletes the records it owns. Use the supplied transaction. Do not start another transaction or call external systems. Throw an error if deletion cannot proceed. An error must roll back all database changes in the deletion transaction.
Call each module explicitly in dependency order. No module registry is required.
The agent lifecycle module also has a restricted configuration.ts entry point.
It saves configuration and invalidates verification in the caller's transaction.
The lifecycle guide lists its permitted callers.
Ordinary configuration updates use the root lifecycle command.
The agent lifecycle module owns agent creation, pause, resume, and termination. Use its commands for these changes.