agentleFS
Simple-With-UsSimple-With-Us/BotFleet/.claude/skills/codex-triage/SKILL.md

codex-triage

Skill4 starsChanged 10 days agoApache-2.0View on GitHub

Triage unresolved GitHub review threads (chatgpt-codex-connector, Cursor Bugbot, and any other review bot or human) — classify against current HEAD, fix real findings in one batch, reply, then resolve. Use when gating a merge, when branch protection blocks on conversation resolution, or when a bot re-reviews after push.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not classify from the stale diff hunk.

    line 29
  2. Distinctive:

    Do not resolve without a reply.

    line 37
  3. Distinctive:

    Never blind-resolve to force a merge.

    line 51
  4. Distinctive:

    If NOT on main, open a follow-up PR from the same branch.

    line 65
  5. Very specific:

    Round 2: only for threads opened on the round-1 head (group by comments.nodes[0].commit.oid).

    line 75
  6. Distinctive:

    Stop: after round 2, do not chase convergence.

    line 78

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Review-thread triage (Universal)

Name is historical (codex-connector). Apply to every unresolved thread on the PR: Codex, Cursor Bugbot, Copilot, humans.

Unresolved threads block merge on every protected fleet repo even when checks are green.

1. Fetch unresolved threads

OWNER="$(gh repo view --json owner --jq .owner.login)"
REPO="$(gh repo view --json name --jq .name)"
gh api graphql -f query="query { repository(owner:\"$OWNER\", name:\"$REPO\") { pullRequest(number: <N>) { reviewThreads(first: 100) { nodes { id isResolved path line comments(first: 10) { nodes { author { login } body diffHunk } } } } } } }" \
  --jq '.data.repository.pullRequest.reviewThreads.nodes | map(select(.isResolved==false))'

2. Classify against current HEAD

Three buckets only:

  • addressed — already fixed on this branch after the comment.
  • false_positive — bot misread context, or the rule does not apply here.
  • real — valid. For money-path / auth / execution / policy / billing files, have a second agent (or a frontier child) adversarially verify before you change behavior.

Do not classify from the stale diff hunk. Read the file at HEAD.

3. Fix reals in one batch

One commit with a regression test per behavior change. Push before resolving any thread (race in step 5).

4. Reply, then resolve

Do not resolve without a reply.

gh api graphql -f query='mutation($t:ID!,$b:String!){addPullRequestReviewThreadReply(input:{pullRequestReviewThreadId:$t,body:$b}){comment{id}}}' -F t=<threadId> -f b=<text>

gh api graphql -f query='mutation($t:ID!){resolveReviewThread(input:{threadId:$t}){thread{isResolved}}}' -F t=<threadId>

Reply shape (review-thread replies are PR text: a real U+00A0 plus a space between sentences, converted with the perl recipe in sentence-gap; since the 2026-10-23 owner ruling, prose files take the same U+00A0 plus a space, and code never does):

  • Real: Fixed in <short-sha>. <function>. Test: test/foo.test.ts.
  • False positive: why the rule does not apply, then resolve.
  • Addressed already: cite the SHA that landed the fix.

Never blind-resolve to force a merge. The gate exists because some findings are real (commit-author, licenses, money-path).

5. Auto-merge race

The instant the last thread resolves and CI is green, auto-merge fires. Bots re-review every push and often never converge.

  • Resolving the last thread is merging the PR. Triple-check first.
  • Round-2 comments often land on an already-merged PR. Before resolving as "fixed," confirm the fix reached main:
git fetch origin
git merge-base --is-ancestor <fix-sha> origin/main && echo "on main" || echo "NOT on main"

If NOT on main, open a follow-up PR from the same branch. Expect squash-merge conflicts — merge main in; this branch's newer rounds win in its own files.

6. Per-PR stop rule (rounds 1–2, then accepted risk)

Kody and similar bots re-review the whole diff on every push. Zero-unresolved threads is not a terminating merge bar — each fix push can open a fresh batch (BF-1997).

Round 1: classify all unresolved threads against HEAD; batch-fix real items; reply + resolve addressed and false positives. Push before resolving (section 5).

Round 2: only for threads opened on the round-1 head (group by comments.nodes[0].commit.oid). Same batch-fix discipline.

Stop: after round 2, do not chase convergence. For each remaining thread, either merge with an explicit accepted risk note (why the finding is deferred, already mitigated elsewhere, or out of scope) or escalate money-path / release-path items to the owner. Post accepted-risk summaries on the PR and, for fleet work, THE BOARD or Linear (BF-1997 canon).

Kody findings are advisory — verify against HEAD; adversarially check critical paths before changing behavior (section 2). Later rounds on an already-merged PR are mostly noise: triage genuine hazards; surface the rest via Apple Notes + board comment.

Full decision: docs/decisions/2026-10-15-kody-review-stop-rule.md.

Canon

  • /Users/jay/apps/FLEET-PROTOCOL.md — Merge requirements
  • App AGENTS.md — Kody Review Rules; Pre-Commit / Handoff
  • docs/decisions/2026-10-15-kody-review-stop-rule.md — stop rule evidence
  • Skills: unstick-pr, land-lane

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.