agentleFS
XakkiXakki/ai-agents-skills/skills/quality-change-review/SKILL.md

quality-change-review

Skill6 starsChanged 3 months agoMITView on GitHub

Independently review or accept a candidate change and verify defect claims with evidence, separating findings, gaps and hypotheses. Use for review or acceptance decisions, not implementation. RU: ревью, проверка приёмки, проверка находки, мутационное тестирование.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If a test failure may disclose sensitive values and checking is authorized, exercise the failing assertion with a synthetic canary and …

    line 19

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Independent review and verified findings

Use for an independent review or acceptance decision, not implementation. Establish the request, review authority, required independence, exact candidate and relevant boundaries before inspecting. Record the current revision, staged/unstaged and relevant untracked work where authorized. When the tree may move, check candidate identity before relying on earlier evidence; a snapshot is not proof of runtime behavior.

Trace the actual behavior from producer and inputs through consumers, state transitions and observable effects. For cross-layer changes inspect shared wiring and the complete integrated range, not only each child verdict. For a prior finding, verify the current candidate and relevant regression; previous reviewer or implementer summaries are leads, not fresh results.

Classify observed defects, evidence gaps and hypotheses separately. State the trigger, affected boundary, impact and precise source or evidence. An unavailable required gate can block acceptance without being a proven product defect. Do not infer authorization to edit, stage, move a board card or publish from review alone. Run only permitted checks with side effects understood; preserve partial, failed and unrun results.

Verifying a claim or finding

  • State precisely what was observed and the boundary of the claim. Code inspection, static analysis, live execution and tests establish different facts; never imply an unexercised path or an unavailable test was verified. For a negative claim, define the searched or tested scope and confirm the observation instrument works.
  • Reproduce a reported failure when practical before relying on its predicted cause. Verify a fix against the relevant behavior and the final artifact: reproducing a defect is not evidence that the fix works. If execution is unsafe, unavailable or unauthorized, use the best permitted evidence and state the limit.
  • Use mutation only when it materially tests the claimed guard and is authorized and safe. Check that the intended test was selected, the code ran, the fixture could express the defect and a failing result failed for the intended reason. Restore only the exact owned change, verify the final bytes and any rebuilt artifact, and preserve unrelated work. Do not mutate in a read-only review.
  • If a test failure may disclose sensitive values and checking is authorized, exercise the failing assertion with a synthetic canary and inspect the retained output without revealing the canary; assertion frameworks can print subexpression values.
  • Counts, searches and output must cover the actual question: truncation, deselection, collection failure and zero records can look cleaner than reality. Report command, scope, outcome and limits; never report a partial or still-running check as passed. Worked examples: instances.

For requested negative guarantees consult negative contract review; for backend reviews backend boundaries; for cross-runtime configuration and session changes integration boundaries. For exact final validation results use quality-change-verify, and for noisy command evidence agent-command-run. Give the requested verdict with actionable findings, evidence and limitations, without overstating untested behavior.

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.