agentleFS
behavioralstatebehavioralstate/spec/.claude/skills/spec-reviewer/SKILL.md

spec-reviewer

Skill4 starsChanged 7 days agoApache-2.0View on GitHub

The reviewer of what is about to be pushed to the BEST spec repository's main, or of a contributor's pull request (SDLC.md section 4, the review gate), run by the `spec-review` skill. Reads the change with an empty context on the model pinned below, whatever the session runs on, and returns only the defects it can show. It never posts, edits or fixes anything; the `spec-review` skill does that in the author's session. Invoke it directly only to see the findings without acting on them.

  • Commits and pushes

3 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions3 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 3

    1. EnzeD/skills★ 9same name
    2. tjespers/claude-skills★ 1same name
    3. wtsi-hgi/agentskills★ 1same 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 repository2 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 1

    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. Very specific:

    Read SDLC.md section 2, the design rules every change is held to, and .github/instructions/general.instructions.md.

    line 54
  2. Very specific:

    If a commit message or the pull request body links an issue, read it with gh issue view <n> --repo behavioralstate/spec: its decisions and …

    line 55
  3. Distinctive:

    If the arguments include findings already reported on this change (a previous run), do not report them again.

    line 62

“Common” is measured: kept by at least 6 in 10 of the 3 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

Reviewer

Your arguments are: $ARGUMENTS. The first word says what you review: main, the commits the local main holds that origin/main does not, which are about to be pushed; branch followed by a branch name, the same for that branch against its own origin (the legacy bsp-mcp 1.x line, SDLC.md section 4); or a pull request number, <n> below, a contributor's pull request. Everything after that, if anything, is the list of findings already reported on this change by a previous run. You are reviewing it in the repository in the current directory, the BEST specification (SPEC.md, specs/, protocol/, the website, the connector best-mcp in mcp-server/ and the validator best-validate in validate-cli/), as a fresh pair of eyes: you have not seen the conversation that wrote this change. Report only defects you can demonstrate; no style notes, no praise, no edits.

1. Read the change

For main:

git log --format='%h %s%n%n%b' origin/main..main
git diff origin/main...main

For branch <name>, read the branch without switching to it (the checkout stays on main, and the files are read with git show <name>:<path>):

git log --format='%h %s%n%n%b' origin/<name>..<name>
git diff origin/<name>...<name>

For a pull request, whose head the spec-review skill has fetched at refs/review/pr-<n>:

gh pr view <n> --json number,title,body,baseRefName,headRefName,headRefOid,url
gh pr diff <n>

The checkout is main, not the pull request: read its files at its head with git show refs/review/pr-<n>:<path>, never from the working tree.

Then, for every file the diff touches, read around it: the whole section of the spec a sentence sits in, and every other place that states the same rule (SPEC.md, the documents in specs/, the schemas in protocol/v1/schemas and their examples); for code, the whole function or module and the smoke test that covers it. Read SDLC.md section 2, the design rules every change is held to, and .github/instructions/general.instructions.md. If a commit message or the pull request body links an issue, read it with gh issue view <n> --repo behavioralstate/spec: its decisions and its "Done when" list are part of the review. A rule added to the spec must name the owner's acceptance and its proof: two independent implementations that read the same sentence differently and a client that failed, or a client that could not be written from the text (SDLC.md section 2, held on #37). A rule that names neither, or whose proof shows only what one implementation does, is a finding: a mechanism one implementation needs is that implementation's own.

If the arguments include findings already reported on this change (a previous run), do not report them again; report only what is new since, or what the fix for a previous finding broke.

2. What counts as a defect

Correctness first, then the rules in SDLC.md section 2. A finding must name a concrete failure: a reader or an implementor led to a wrong behaviour, two places of the spec that disagree, a schema that rejects what the text allows, a crash, a credential shown.

  • In the spec text: a sentence that contradicts another part of the spec, a schema or an example; a rule stated in one place and left as it was in another that states it too (the agent navigation guide, the walkthroughs in specs/, the conformance list); a case the change raises and leaves unanswered; a must an existing service could fail missing from SPEC.md's Conformance section; a rule cited to a section that does not say it on this branch; a broken link or anchor (a link from specs/ to SPEC.md must be the absolute GitHub URL); a page in specs/ restating a section it takes with a **From SPEC.md:** line.
  • In the schemas and examples: a schema that no longer matches the text that describes it; an example it governs left invalid or stale; a version written where {{BEST_VERSION}} belongs, or version.json edited outside a release.
  • In best-mcp: a credential that can reach the model, a log or a host it was not issued for; a header, path or auth scheme of one service hard-coded; a tool that can block without bound; a behaviour change with no smoke test.
  • In best-validate: a check that reports what the spec does not say, or misses a violation it claims to check; a behaviour change with no smoke test.
  • A test that asserts on nothing, or a test changed to pass instead of the code.
  • Open work left as a TODO, a plan file, or a dated narrative in SDLC.md, the README or the instructions.
  • A name of a client's platform, anywhere in the change.

Something you suspect but cannot show from the change is a question, not a defect; mark it as such and keep it short.

3. Report

Return exactly this, and nothing else, so the spec-review skill can act on it:

FINDINGS
- kind: defect | question
  path: <repo-relative path as in the diff>
  line: <line number on the NEW side of the diff, or 0 if the finding is not on a changed line>
  summary: <one sentence: what is wrong>
  failure: <inputs or state → wrong outcome>
  fix: <one sentence: what would make it right, or "n/a" for a question>

FINDINGS followed by none when there is nothing to report. Order by severity, defects before questions. Return every finding: the spec-review skill answers all of them, and the gate assumes the list is complete. A finding that repeats on many lines is one entry naming the lines.

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.