agentleFS
guimatheus92guimatheus92/pr-review/.claude/skills/acceptance/SKILL.md

acceptance

Skill3 starsChanged 31 days agoMITView on GitHub

Test a PR review end to end and prove it honoured every guarantee in INVARIANTS.md. Use when asked to test/validate a review of a pull request, to check a finished run, or to run the 3-provider x 2-runtime acceptance matrix. Drives `pr-review verify` and `scripts/acceptance.mjs` — never hand-checks the guarantees.

  • Reads credentials

7 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions7 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 7

    1. lobehub/lobe-chat★ 86ksame name
    2. lobehub/lobehub★ 86ksame name
    3. smixs/code-quality★ 16same name
    4. r0shi/harborclerk★ 3same name
    5. Daeguk-Sun/dcNess★ 1same name
    6. ofweb/skills★ 1same name
    7. Freakling/Director-Driven-Development★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository4 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not hand-check the guarantees: If you find yourself reading a PR in the browser to see whether a comment is inline, you have left the …

    line 13
  2. Distinctive:

    Run these in order. Do not skip a step because the previous one looked fine.

    line 19
  3. Very specific:

    verify never posts, deletes or edits anything: If a step is about to change the PR, that step is not verify.

    line 89
  4. Very specific:

    A verify FAIL on a PR that advanced since the run: should already be a SKIP with the SHAs named.

    line 98

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “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

Testing a PR review

The checklist is a program, not prose. pr-review verify renders one row per invariant in INVARIANTS.md, always the full list, and exits 2 on any FAIL — so there is no version of this task where an item gets skipped by accident. Your job is the judgment: read the FAIL rows, find out why, report honestly.

Do not hand-check the guarantees. If you find yourself reading a PR in the browser to see whether a comment is inline, you have left the deterministic path. Fix the check instead.

Steps

Run these in order. Do not skip a step because the previous one looked fine.

  1. Build. npm run build verify and the matrix drive dist/cli.cjs, not src/. A stale bundle audits code that is not the code under test.
  2. Review the PR.
    node dist/cli.cjs review <pr-url>            # posts
    node dist/cli.cjs review <pr-url> --dry-run  # previews
    Long runs: add --detach and poll node dist/cli.cjs status <run-id>.
  3. Audit it.
    node dist/cli.cjs verify --pr <pr-url>
    Exit 0 = every invariant PASS or SKIP. Exit 1 = the audit could not be completed (the PR read-back failed) — report that as "not verified", never as "clean". Exit 2 = at least one FAIL, and a FAIL is a product defect until proven otherwise. --offline skips the live PR read on purpose (every row that needs it reports SKIP, and the exit stays 0); --json for machine use.
  4. The matrix — all three providers, every time.

    npm run acceptance                    # 3 providers x 2 runtimes + the file-list and no-fetch gates

    GitHub alone is not a result. Azure DevOps and GitLab have full providers and zero non-stubbed coverage anywhere else in this repo, so a matrix narrowed to --provider github proves the one path that was already proven. Narrow it only to re-run a single failing cell while fixing it, and re-run the whole thing before you call the work done.

    npm run acceptance -- --provider gitlab --runtime claude   # iterating on one cell

    Credentials resolve per provider as env var → .env → that provider's own CLI login (gh auth token, az account get-access-token, glab config get token), so the same command works on a laptop and in CI. A missing one throws naming exactly what to set or which CLI to log into — that is a setup error to fix, never a provider to drop from the run.

    See evals/acceptance/README.md — it posts to real fixture PRs and needs the estate seeded first (npm run acceptance:seed).

  5. Report. Give the verify table verbatim. Every FAIL with its evidence, every SKIP with its reason. Never summarise the table into a sentence and never drop a row — a missing row reads as "fine", which is the exact failure this whole mechanism exists to prevent.

Reading the result

RowMeans
PASSChecked against evidence and held.
FAILViolated, or a check threw. Either way, investigate — never explain it away.
SKIPNot checkable and it says why: a dry run, --offline, a PR that moved on, or an invariant only the test suite can guard.

A run with many SKIPs is not a clean run. If INV-POST-* are all SKIP, nothing was posted and nothing about posting was proven.

The matrix report adds two more, and neither is a pass: ⏭️ skip means the cell was not requested, 🚧 blocked means it could not run (today: Copilot with its premium requests spent). Both are counted apart from the passes. Read 3 cell(s) passed, 3 blocked as three providers proven on one runtime — never as a green matrix.

Gotchas

  • verify never posts, deletes or edits anything. If a step is about to change the PR, that step is not verify.
  • A --dry-run review cannot prove the posting invariants. They report SKIP by design. Asking "does it post correctly?" needs a real publish run.
  • The acceptance matrix posts for real. It resets the fixture PR before each cell — never after, because that is the step a cancelled run skips.
  • The claude cells do not run in CI (no Anthropic credential there, by choice). They run locally. The CI report marks them skipped rather than omitting them; do not read a green CI run as "both runtimes covered".
  • A verify FAIL on a PR that advanced since the run should already be a SKIP with the SHAs named. If it is a FAIL instead, that is a bug in the check.

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.