fix(skills): drop the residual-findings record file for real sinks

[tmchow]

Summary

Accepted review findings had three possible sinks: a Known Residuals section in the PR description, a ticket in the project's tracker, and — last resort in ce-work and ce-debug, primary in lfg — a committed <root>/residual-review-findings/<branch-or-sha>.md file. This removes the third.

The evidence against it is one-sided. It fired once in this repository's history, and that once was wrong: a tracker ticket existed and was filed, and the file was written anyway, though it is specified as reachable only when no tracker is. That issue has since closed and the file was still sitting in the tree, because nothing ever collects one. A path named after a branch or SHA in a directory nobody greps is closer to write-only than to durable.

It also cost more than it carried. Every consumer had to describe it, and it forced an exception into ce-debug's commit-scope constraint — "the fix-owned files and nothing else, except this one artifact" — which was itself a contract break caught in review on #1385, since the artifact is not a fix-owned file.

Where residuals go now

Situation	Sink
A PR will be created or updated	Known Residuals section in the PR description
No PR, tracker reachable	One ticket per finding
lfg — findings its tracker chain could not file (failed / no_sink)	One run-report comment on the PR
Neither a PR nor a tracker	Stated in the run report, explicitly as recorded nowhere else

The lfg change is the substantive one. Its DONE gate previously blocked on the record file being written, committed, and pushed. It now blocks on the run-report comment — the same surface ce-babysit-pr already uses for unfixable CI, so this consolidates onto an existing pattern rather than inventing one. A point-in-time comment also does not go stale as items resolve, which is the property lfg cited when it rejected a PR-body section.

The last row is the honest part. With no PR and no tracker there is no durable sink, and saying so is better than writing a file that only looks like one. That case is also the narrowest: it needs a repo with neither a remote nor a tracker, where the run output is in front of the user anyway.

Scope

Removed from ce-work, ce-debug (skill, pipeline-mode, post-fix-handoff), lfg, the ce-doc-review protected-artifact list, the config template and its byte-identical config.example.yaml copy, and the user-facing docs/skills/lfg.md. The stale artifact is deleted.

tests/docs-root-literals.test.ts deliberately keeps residual-review-findings in its subdirectory list: that guard catches hardcoded docs/<subdir> literals, and keeping the name means a reintroduced literal would still fail rather than pass silently.

Historical planning docs under docs/plans/ still mention the mechanism and are left alone — they are records of what was decided at the time.

Validation

bun test 3101 pass / 0 fail; release:validate and plugin:validate --strict clean. Two contract tests in tests/review-skill-contract.test.ts pinned the removed path and the record file's push mechanics; both now pin the replacement behaviour instead.

Related: #1385

Security Disclosure

No security-relevant changes. This removes a write: lfg no longer creates, commits, and pushes a file as part of its DONE gate, so the autonomous pipeline touches the tree less than before.

Agent Disclosure

• Model: Claude Code · claude-opus-5[1m]

[cursor]

PR Summary

^{Cursor Bugbot is generating a summary for commit 7db6f4b. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 7db6f4b. Configure here.}

[cursor]

PR comment posted before PR exists

High Severity

Step 6 now makes residuals durable by posting a run-report comment with gh pr comment, but the PR is only created later in step 8 via ce-commit-push-pr. On a fresh branch the comment target does not exist yet, so the new DONE gate either stalls or skips the comment, leaving failed / no_sink findings without the durable sink this change intended. Step 8 back-fills ticket links once the PR URL is known, but never posts a deferred residual comment.

Additional Locations (1)

• skills/lfg/SKILL.md#L112-L115

 

^{Reviewed by Cursor Bugbot for commit 7db6f4b. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7db6f4bf48

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Delay the residual comment until after PR creation

On the normal LFG path where the current branch has no existing PR, this step runs before step 8 opens one. The checked gh pr comment --help defines the command as adding a comment to an existing pull request, so it has no target here; no later step retries the composed comment after receiving the PR URL, leaving failed and no_sink findings without the promised durable record. Carry the section forward and post it immediately after step 8 creates or identifies the PR.

AGENTS.md reference: AGENTS.md:L147-L147

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Remove the stale claim that residuals are committed

After this changed summary establishes tracker tickets plus a PR comment as the residual sinks, the same user-facing page still says at line 106 that unapplied findings are “filed and committed.” The record-file commit was removed by this change, so that description now falsely promises a committed artifact; update it to describe the tracker/comment behavior consistently.

AGENTS.md reference: AGENTS.md:L88-L90

Useful? React with 👍 / 👎.