mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-27 11:20:42 +01:00
4394fcd00db3eef8d75fbe5cf3740dc059757ea2
5
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
71a8464fee |
[codex] prevent invalid agents from receiving assignments and runs (#7663)
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - The control plane owns agent lifecycle, issue assignment, routine dispatch, heartbeat wakeups, and recovery paths > - Terminated, paused, pending-approval, or otherwise invalid agents should not receive new work or new execution attempts > - The old behavior left eligibility checks spread across routes and services, so assignment and run paths could drift apart > - This pull request centralizes agent lifecycle eligibility and applies it consistently to assignment, invocation, routines, recovery, and UI affordances > - The benefit is safer autonomy: terminated agents stay paused, invalid org-chain agents are surfaced, and active agents keep receiving valid work ## Linked Issues or Issue Description Refs #5103 Related: #1864 Bug fix context: - What happened: agent assignment and heartbeat/run paths did not share one eligibility contract, so invalid lifecycle states could still be considered in some paths. - Expected behavior: terminated agents must never receive new assignments or heartbeat runs, and paused or otherwise invalid agents should be treated as non-invokable consistently. - Steps to reproduce: create or select an agent in an invalid lifecycle state, then attempt assignment, routine dispatch, or heartbeat/recovery wake paths. - Paperclip version/commit: fixed on top of `paperclipai/paperclip` `master` at the PR base. - Deployment mode: applies to the server control plane in local and authenticated deployments. ## What Changed - Added shared agent lifecycle eligibility helpers and exported the related shared types. - Centralized server-side assignability and invokability checks for issue assignment, agent routes, heartbeat dispatch, routines, recovery, and liveness logic. - Hardened issue assignment so invalid assignees are rejected instead of queued for work. - Hardened heartbeat/routine/recovery paths so terminated and otherwise invalid agents are not woken for new runs. - Updated board UI affordances to disable invalid agent actions and surface org-chain warnings where relevant. - Added targeted shared, server, and UI tests for the new eligibility behavior. ## Verification - `pnpm exec vitest run packages/shared/src/agent-eligibility.test.ts server/src/__tests__/agent-invokability.test.ts server/src/__tests__/heartbeat-archived-company-guard.test.ts server/src/__tests__/issue-liveness.test.ts server/src/__tests__/issues-service.test.ts server/src/__tests__/routines-service.test.ts ui/src/lib/company-members.test.ts ui/src/pages/Agents.test.tsx` — 8 files, 144 tests passed. - `pnpm --filter @paperclipai/shared typecheck && pnpm --filter @paperclipai/server typecheck && pnpm --filter @paperclipai/ui typecheck` — passed. - Checked the PR diff does not include `pnpm-lock.yaml` or `.github/workflows` changes. - Checked `ROADMAP.md`; this is a targeted control-plane safety fix and does not duplicate a planned core feature. - Searched GitHub for duplicate or related PRs/issues; closest related items are linked above. - CI and Greptile verification are pending on the opened PR and will be followed up before requesting merge. ## Risks Low to moderate risk. The intended behavioral shift is that invalid agents are refused earlier and more consistently, which could expose existing data with paused, pending, terminated, or broken org-chain assignees. The added tests cover the critical assignment, heartbeat, routine, recovery, shared helper, and UI paths. No database migrations are included. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used OpenAI GPT-5 Codex via the Paperclip `codex_local` adapter, with shell/git/GitHub CLI tool use. Reasoning mode and context window are managed by the adapter runtime and 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 run tests locally and they pass - [x] I have added or updated tests where applicable - [x] If this change affects the UI, I have included before/after screenshots (not applicable: no design screenshots requested; UI behavior is covered by tests) - [x] I have updated relevant documentation to reflect my changes (not applicable: no user-facing command or schema docs changed) - [x] I have considered and documented any risks above - [ ] All Paperclip CI gates are green (pending CI) - [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups (pending Greptile) - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing> |
||
|
|
e400315cbf |
Guard assigned backlog liveness (#5428)
## Thinking Path > - Paperclip orchestrates AI agents for zero-human companies > - The issue graph and liveness recovery system decide whether assigned work is executable or parked > - Assigned issues created without an explicit status could silently land in backlog, making parents look blocked with no productive wake path > - The server, shared validators, recovery analysis, and UI all need to agree on that execution semantic > - This pull request makes assigned issue creation default to `todo`, flags assigned backlog blockers, and surfaces the state in the board > - The benefit is that parked assigned work becomes intentional and visible instead of creating silent liveness stalls ## What Changed - Adds contract tests for assigned issue creation defaults. - Defaults assigned issue creation to `todo` when status is omitted while preserving explicit `backlog` parking. - Exposes `resolveCreateIssueStatusDefault` through shared validators. - Teaches liveness/blocker attention paths to distinguish assigned backlog blockers. - Adds UI notices, row/header badges, and issue detail safeguards for assigned backlog blockers. - Adds Storybook fixtures and execution-semantics documentation for the assigned-backlog behavior. ## Verification - `pnpm run preflight:workspace-links && pnpm exec vitest run packages/shared/src/validators/issue.test.ts server/src/__tests__/issue-assigned-backlog-contract-routes.test.ts server/src/__tests__/issue-blocker-attention.test.ts server/src/__tests__/issue-liveness.test.ts server/src/__tests__/heartbeat-issue-liveness-escalation.test.ts ui/src/components/IssueAssignedBacklogNotice.test.tsx ui/src/components/IssueRow.test.tsx` — 50 passed, 23 skipped. - Skipped tests were embedded Postgres suites on this host with the repo skip message: `Postgres init script exited with code null. Please check the logs for extra info. The data directory might already exist.` - Pairwise merge check against the issue-controls PR branch completed without conflicts via `git merge --no-commit --no-ff` in a temporary worktree. - Screenshots for assigned-backlog UI states: [light](docs/pr-screenshots/pr-5428/assigned-backlog-light.png), [dark](docs/pr-screenshots/pr-5428/assigned-backlog-dark.png). - Follow-up checks: `pnpm --filter /ui typecheck`; `pnpm --filter /mcp-server build`; `pnpm --filter /mcp-server test`; `pnpm exec vitest run packages/shared/src/validators/issue.test.ts`; focused UI component tests. - Remote PR checks on head `6300b3c`: policy, verify, serialized server shards 1/4-4/4, Canary Dry Run, e2e, Greptile Review, and Snyk all passed. ## Risks - Medium: changes status defaulting for assigned issue creation when the caller omits status. Explicit `backlog` remains supported, and server/shared tests cover both paths. - Medium: liveness classification changes can affect blocker attention labels; focused service and UI tests cover the new assigned-backlog state. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - OpenAI Codex coding agent, GPT-5 model family (`gpt-5`), tool-enabled Paperclip heartbeat environment. Context window and internal reasoning mode are not exposed by the runtime. ## 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 run tests locally and they pass - [x] I have added or updated tests where applicable - [x] If this change affects the UI, I have included before/after screenshots - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing> |
||
|
|
82e257c7ba |
Cancel stale queued heartbeats when issue graph changes (PAP-2314) (#4534)
Co-Authored-By: Paperclip <noreply@paperclip.ing> |
||
|
|
5a0c1979cf | [codex] Add runtime lifecycle recovery and live issue visibility (#4419) | ||
|
|
1954eb3048 |
[codex] Detect issue graph liveness deadlocks (#4209)
## Thinking Path > - Paperclip orchestrates AI agents for zero-human companies. > - The heartbeat harness is responsible for waking agents, reconciling issue state, and keeping execution moving. > - Some dependency graphs can become live-locks when a blocked issue depends on an unassigned, cancelled, or otherwise uninvokable issue. > - Review and approval stages can also stall when the recorded participant can no longer be resolved. > - This pull request adds issue graph liveness classification plus heartbeat reconciliation that creates durable escalation work for those cases. > - The benefit is that harness-level deadlocks become visible, assigned, logged, and recoverable instead of silently leaving task sequences blocked. ## What Changed - Added an issue graph liveness classifier for blocked dependency and invalid review participant states. - Added heartbeat reconciliation that creates one stable escalation issue per liveness incident, links it as a blocker, comments on the affected issue, wakes the recommended owner, and logs activity. - Wired startup and periodic server reconciliation for issue graph liveness incidents. - Added focused tests for classifier behavior, heartbeat escalation creation/deduplication, and queued dependency wake promotion. - Fixed queued issue wakes so a coalesced wake re-runs queue selection, allowing dependency-unblocked work to start immediately. ## Verification - `pnpm exec vitest run server/src/__tests__/heartbeat-dependency-scheduling.test.ts server/src/__tests__/issue-liveness.test.ts server/src/__tests__/heartbeat-issue-liveness-escalation.test.ts` - Passed locally: `server/src/__tests__/issue-liveness.test.ts` (5 tests) - Skipped locally: embedded Postgres suites because optional package `@embedded-postgres/darwin-x64` is not installed on this host - `pnpm --filter @paperclipai/server typecheck` - `git diff --check` - Greptile review loop: ran 3 times as requested; the final Greptile-reviewed head `0a864eab` had 0 comments and all Greptile threads were resolved. Later commits are CI/test-stability fixes after the requested max Greptile pass count. - GitHub PR checks on head `87493ed4`: `policy`, `verify`, `e2e`, and `security/snyk (cryppadotta)` all passed. ## Risks - Moderate operational risk: the reconciler creates escalation issues automatically, so incorrect classification could create noise. Stable incident keys and deduplication limit repeated escalation. - Low schema risk: this uses existing issue, relation, comment, wake, and activity log tables with no migration. - No UI screenshots included because this change is server-side harness behavior only. > For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and discuss it in `#dev` before opening the PR. Feature PRs that overlap with planned core work may need to be redirected — check the roadmap first. See `CONTRIBUTING.md`. ## Model Used - OpenAI Codex, GPT-5-based coding agent. Exact runtime model ID and context window were not exposed in this session. Used tool execution for git, tests, typecheck, Greptile review handling, and GitHub CLI operations. ## 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 run tests locally and they pass - [x] I have added or updated tests where applicable - [x] If this change affects the UI, I have included before/after screenshots - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] I will address all Greptile and reviewer comments before requesting merge |