agentleFS
isac7722isac7722/skills/check-review/SKILL.md

check-review

Skill1 starsChanged 5 months agoMITView on GitHub

Verify reviewer comments against the codebase, separate valid findings from incorrect or non-applicable ones, and produce a developer-confirmed fix plan. Use this skill when the user says "/check-review", provides reviewer comments separated by "---", asks to validate review feedback against code, or wants a confirmation step before applying reviewer-requested changes. Do not edit code before explicit developer confirmation.

The file

Show its rules and text

We split this skill into the rules it gives an agent

20 rules. About 0 are common, 17 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈17

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not accept reviewer comments at face value.

    line 10
  2. Very specific:

    When the user provides review feedback with /check-review, process it as follows:

    line 14
  3. Distinctive:

    If no delimiter is present, temporarily split by paragraphs, numbered lists, or bullets, and tell the user what split rule was used.

    line 18
  4. Distinctive:

    If a review comment is too vague to locate code or infer intent, do not modify anything.

    line 19
  5. Distinctive:

    Before editing code, collect the relevant context:

    line 25
  6. Distinctive:

    Run git status --short to identify existing user or working-tree changes.

    line 27
  7. Very specific:

    Use rg and file navigation to find the code, tests, configuration, or docs referenced by the review.

    line 28
  8. Distinctive:

    When useful, inspect git diff, git log, tests, and call sites to trace current behavior.

    line 29
  9. Distinctive:

    Do not treat the reviewer's wording as evidence by itself.

    line 30
  10. Very specific:

    Do not force uncertain items into Valid.

    line 42
  11. Distinctive:

    Use clickable file paths with line numbers when possible.

    line 72
  12. Distinctive:

    Do not edit code, tests, docs, or configuration before final developer confirmation.

    line 76
  13. Distinctive:

    Treat these as explicit confirmation:

    line 78
  14. Distinctive:

    Do not treat these as confirmation:

    line 86
  15. Distinctive:

    If the user approves only some items, modify only those approved items.

    line 93
  16. Distinctive:

    Do not revert or overwrite unrelated user changes.

    line 100
  17. Distinctive:

    Run relevant tests, type checks, or linters when available.

    line 101
  18. Distinctive:

    If verification cannot be run, state why.

    line 102
  19. Distinctive:

    Prefer code evidence over reviewer claims when they conflict.

    line 108
  20. Distinctive:

    Do not repeat review comments at length.

    line 110

≈ “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

Check Review

Overview

Do not accept reviewer comments at face value. Check the current codebase, determine whether each comment is factually supported, and include only evidence-backed items in the proposed fix plan. Do not change code until the developer gives explicit final confirmation.

Input Parsing

When the user provides review feedback with /check-review, process it as follows:

  • Split review blocks on lines containing ---.
  • Preserve the intent of each original block, but split a block into sub-items when it contains multiple independently verifiable claims.
  • If no delimiter is present, temporarily split by paragraphs, numbered lists, or bullets, and tell the user what split rule was used.
  • If a review comment is too vague to locate code or infer intent, do not modify anything. Ask only the 1-3 questions required to proceed.

Workflow

1. Gather Context

Before editing code, collect the relevant context:

  • Run git status --short to identify existing user or working-tree changes.
  • Use rg and file navigation to find the code, tests, configuration, or docs referenced by the review.
  • When useful, inspect git diff, git log, tests, and call sites to trace current behavior.
  • Do not treat the reviewer's wording as evidence by itself. Establish file, line, and behavior evidence from the repository.

2. Verify Each Review Item

Classify each review item into one of these verdicts:

  • Valid: The code evidence supports the review and a change is needed.
  • Already handled: The requested behavior already exists or is already covered by the current changes.
  • Incorrect / Not applicable: The code evidence contradicts the review, or the comment is outside the current scope.
  • Needs clarification: The intent or code location is unclear enough that a reliable verdict is not possible.
  • Optional / Trade-off: The comment is a style, structure, or operational preference rather than a correctness issue.

Do not force uncertain items into Valid. Say no when the review is wrong or not applicable, and explain why using code evidence.

3. Report Findings Before Planning

Share the verification results before making any code change. Use this default structure:

## Review Check

### Verdicts

| # | Verdict | Review Summary | Code Evidence | Reason |
|---|---------|----------------|---------------|--------|
| 1 | Valid | ... | `path/file.ext:123` | ... |
| 2 | Incorrect / Not applicable | ... | `path/file.ext:45` | ... |

### Proposed Plan

1. ...
2. ...

### Not Applying

- ...

### Confirmation Needed

Should I proceed with the changes in this plan?

Use clickable file paths with line numbers when possible.

4. Require Developer Confirmation

Do not edit code, tests, docs, or configuration before final developer confirmation.

Treat these as explicit confirmation:

  • "Proceed"
  • "Apply the changes"
  • "Fix it"
  • "Apply items 1 and 3 only"
  • "Go with the plan"

Do not treat these as confirmation:

  • "What do you think?"
  • "Would that work?"
  • "Review it more"
  • "Show me the plan"

If the user approves only some items, modify only those approved items. Leave unapproved items unchanged.

5. Apply Approved Changes

After confirmation only:

  1. Make the smallest practical code changes for the approved review items.
  2. Do not revert or overwrite unrelated user changes.
  3. Run relevant tests, type checks, or linters when available.
  4. If verification cannot be run, state why.
  5. In the final response, briefly summarize applied items, skipped items, and verification results.

Response Rules

  • Reply in the same language the user uses for the active request unless they ask for a specific language.
  • Prefer code evidence over reviewer claims when they conflict.
  • Mark assumptions explicitly and confirm them before putting assumption-dependent work into the plan.
  • Do not repeat review comments at length. Focus on verdicts, evidence, and next actions.
  • Include both a rough implementation direction and concrete files or actions in the plan.

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.