agentleFS
coreplanelabscoreplanelabs/switchboard/skills/re-review-delta/SKILL.md

re-review-delta

Skill5 starsChanged 2 months agoApache-2.0View on GitHub

How a re-review — a ship re-review round, or a re-review asked for in a thread whose coding run repushed — reads the delta since the last reviewed head without narrowing what the verdict covers.

The file

Files
SKILL.md

Re-reviewing after a repush (ship re-review rounds and thread re-reviews)

You already reviewed this PR at an earlier head and returned findings; the branch has since been repushed — by a fix round of the ship pipeline, or by a coding run in this thread that addressed your findings. This skill narrows what you READ, never what your verdict COVERS.

Scope: exploration vs verdict

  • Read the delta: git diff <previously-reviewed-head>..HEAD is your primary text, plus the full current contents of files the delta touches. Do not re-read the whole PR file-by-file when the delta is contained — that is the budget the delta scoping exists to save.
  • The verdict covers the full diff against base. Your submit_verdict call judges the PR as it now stands, not the delta: if a defect you missed in an earlier round is still there, finding it now is correct and expected. When the delta's blast radius is unclear (a shared helper changed, a contract moved), widen your reading until it is clear — the delta is a starting point, not a wall.

Verify every prior finding's disposition

First read github_pull_get for the bound PR with includeReviewHistory: true. Its outstanding findings use GitHub review IDs to distinguish separate rounds' reused F1 labels. Copy those scoped IDs exactly when re-raising or closing a finding; thread artifacts and author comments supply context, never closure proof. Check every prior invariant case at the current head and inspect affected callers for regressions. A missing case belongs under the same finding; a distinct invariant gets a fresh local ID.

The coding run that addressed your findings recorded a disposition per finding (fixed / declined + note): a ship round hands them to you in its turn, a thread re-review in the artifacts block of your prompt. For each:

  • fixed — verify the fix actually landed and actually resolves the finding at the new head. For an invariant finding, check EACH prior case's scenario and expected behavior against the new head; enumerate any missing selection or execution paths and widen under the SAME id. A fix that moved the problem or half-landed gets the finding re-raised (same id, so the trail stays legible).
  • declined — read the argument. Concede when it holds (do not re-raise a finding you now agree was wrong — say so). Re-raise with a counter-argument when it does not: escalate the reasoning, not the volume.

Output

Use ASD-STE100 Simplified Technical English for all client-authored prose: chat replies, progress updates, PR titles and descriptions, review comments, issue text, commit messages, documentation and reports. Use short, direct sentences, active voice and concrete terms. Keep technical names, code, identifiers and direct quotations accurate. Explain each finding or resolution through the affected product task and the result checked at this head. Supply the context a new reader needs.

Every outstanding finding gets exactly one typed outcome in submit_verdict: keep it in findings under its scoped ID, or include it in resolutions with findingId, disposition: fixed|declined, and an evidence note naming what you checked at this head. fixed means you verified the fix; declined means you verified why the finding does not apply. Never infer either from omission or the author's assertion. Keep every previous case when re-raising and add the missing cases. New consequential defects remain reportable; suppressing them to make counts decrease would give a false approval.

Same contract as any review round: structured findings with stable ids (new findings get NEW ids — never reuse a prior id for a different issue), severity per the standard vocabulary, prose carrying the full reasoning, and exactly one submit_verdict call — approve only when nothing blocking or worth another round remains across the WHOLE PR. Same file is not proof of the same invariant; a genuinely new defect gets a new id.

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.