mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-07 07:23:08 +02:00
scotttongandClaude Opus 4.8
fb2b760915
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work > - Task/issue threads render each comment as a chat bubble; the author determines whether it shows as a left-aligned agent bubble (name + icon) or a right-aligned "Board" bubble > - Comments posted by an agent from a local execution environment are written with a non-human author id (`local-board`/system), so they were mis-rendered as blue "Board" bubbles instead of being attributed to the authoring agent > - This misattribution is confusing (it looks like the human board said something an agent actually said) and it can drive false wake/reconciliation behavior on the affected threads > - This pull request adds server-side attribution derivation (lossless run-id join first, then an explicit run-log post marker), persists the derived agent so the read path stops re-scanning run logs, and stops the client from labeling agent-derived comments "Board" > - The benefit is agent comments render as the correct agent, genuine human board comments are never reattributed, and reads get cheaper after a one-time persistence ## Linked Issues or Issue Description <!-- No public GitHub issue — describing the problem in-PR (bug report shape). --> **What happened?** In a task/issue comment thread, comments authored by an agent from a local execution environment are stored with a non-human author id (`local-board`/system). The UI renders these as right-aligned blue "Board" bubbles, implying a human board member authored them. The mislabeling is also a wake/reconciliation hazard: an agent comment that reads as "Board" can look like human board input. **Expected behavior** Such comments should render as the authoring agent (left-aligned bubble with agent name + icon). Genuine human/board comments must continue to render as "Board" and must never be reattributed to an agent. **Steps to reproduce** 1. Have an agent post a comment on an issue from a local execution environment (author id `local-board`). 2. Open the issue comment thread in the UI. 3. Observe the agent's comment rendered as a right-aligned blue "Board" bubble instead of the authoring agent. **Root cause** The read path did not resolve the authoring agent for these comments, and the client fell back to a "Board" label for the `local-board` author. ## What Changed - **Server derivation (`server/src/services/issues.ts`):** - Resolve the authoring agent from the comment's run id first (`createdByRunId`/`derivedCreatedByRunId` → `heartbeatRuns.agentId`) — lossless when present. - Second tier `run_log_comment_post`: read the run log lazily (only for still-unresolved comments) to match the explicit `comment id:` post marker. - **Guard:** never reattribute a comment whose author maps to a genuine user profile. Only the non-human sentinel (`local-board`, which is itself a `user` row) and authors absent from the `user` table are eligible. - Pure timing-overlap tiers are intentionally **not** used (Option A) — an agent comment and a human board comment posted during the same run are indistinguishable rows, so any timing guess risks mislabeling a real human comment. - **Persistence (`packages/db/src/migrations/0126_issue_comment_derived_attribution.sql`, `packages/db/src/schema/issue_comments.ts`):** add stored `derived_*` attribution columns and write the resolved agent back with a single bulk `UPDATE ... FROM (VALUES ...)`, so reads stop recomputing from run logs. Migration is additive (new nullable columns) with a batched, idempotent backfill of the lossless run-id tier over historical rows. - **Types (`packages/shared/src/types/issue.ts`):** expose the persisted attribution fields and the `IssueCommentDerivedAuthorSource` union. - **Client (`ui/src/lib/issue-chat-messages.test.ts`):** the message builder already prefers a resolved agent id (`authorAgentId ?? runAgentId ?? derivedAuthorAgentId`), so once the server persists the derived agent the bubble renders as the agent automatically — no client code change needed. Adds a regression guard confirming a genuine board comment with no derived agent is still rendered as "Board". - **Tests:** derivation + message-building tests, including assertions that genuine board/user comments are **not** reattributed. ## Verification - `cd server && npx vitest run issues-service` — 94 tests pass: run-id resolution, no-attribution on timing overlap alone (Option A), multi-run ambiguity, same-agent multi-run, and the genuine-user guard. Exercises the real persistence path (bulk UPDATE) against the test DB. - `cd ui && npx vitest run issue-chat-messages` — 27 tests pass; client no longer labels agent-derived comments "Board", and a genuine board comment with no derived agent is not re-labeled. - `cd server && npm run typecheck` — passes (exit 0). - Manual: on a thread containing old agent-authored comments, the blue "Board" bubbles render as the authoring agent; a genuine board comment on the same thread still renders as "Board". ## Risks - **Mis-reattributing a genuine board comment made during an agent run** → mitigated by the human-profile guard (only `local-board`/system authors are eligible) and by dropping pure timing tiers (Option A): only the lossless run-id join and the explicit run-log post marker attribute history. - **Backfill volume / run-log reads** → the migration backfill is batched (5000 rows/loop) and results are persisted so reads stop recomputing; the read-path persistence is a single bulk UPDATE rather than per-comment round-trips. Migration adds only nullable columns (no destructive change). - The persistence/backfill has **not** been run against any production database as part of opening this PR. ## Model Used Claude Opus 4.8 (`claude-opus-4-8`), extended thinking, via Claude Code with tool use. ## 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 — related open PRs (#6006 narrow attribution run scan, #4729 attribution roll-up, #7014 reaped-run attribution) address different attribution paths; none fix the `local-board` "Board" bubble rendering this PR targets. Supersedes #8832 (same change; branch renamed to drop an internal ticket id per CONTRIBUTING → Branch Naming) - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change and contains no internal Paperclip ticket id 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: Claude Opus 4.8 <noreply@anthropic.com>