agentleFS
tardigrdetardigrde/agent-skill-eval/skills/review-diff/SKILL.md

review-diff

Skill4 starsChanged 5 months agoMITView on GitHub

Review a proposed code change (a .diff/.patch file) against the current source code and deliver a structured, severity-tagged review verdict in chat, without modifying or creating any files.

19 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use grilling.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • GoogleSkill

    grilling

    Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 9 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 19 edited versions we read keep these.

  • 9

    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:

    Use when the user asks to review a diff, patch, or proposed change before merging.

    line 16
  2. Distinctive:

    Do NOT apply the diff. Do NOT modify any source file.

    line 24
  3. Distinctive:

    Only flag real defects. Do not pad the review with style nits, speculation, or restatements of what the diff does.

    line 28
  4. Distinctive:

    Read the diff file the user pointed at.

    line 34
  5. Distinctive:

    Read the current version of every file the diff touches, so you understand the context the change lands in.

    line 38
  6. Very specific:

    Check claimed behavior against actual behavior: if the docstring says "threshold or more", the code must use >=.

    line 48
  7. Distinctive:

    All of these must be true:

    line 75
  8. Distinctive:

    No source file was modified and no new file was created

    line 78
  9. Distinctive:

    If the diff file does not exist, say which file was expected and stop

    line 83
  10. Distinctive:

    If the diff does not apply cleanly to the current source (context mismatch), report that as a HIGH finding

    line 84
  11. Very specific:

    If the diff is empty, reply with "No findings." and VERDICT: APPROVE

    line 85

“Common” is measured: kept by at least 6 in 10 of the 19 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

What I do

Review a proposed code change — provided as a unified diff file — against the current source code in the workspace. I read both, look for real defects, and deliver a structured review directly in my reply. I am strictly read-only: I never apply the diff, never edit source files, and never write report files.

When to use me

Use when the user asks to review a diff, patch, or proposed change before merging. Trigger on phrases like:

  • "review this diff"
  • "can you review changes.diff before I merge?"
  • "is this patch safe to apply?"
  • "look over this proposed change"

IMPORTANT: Read-only

Do NOT apply the diff. Do NOT modify any source file. Do NOT create any new files (no report files, no notes). The entire deliverable is the review text in your reply.

IMPORTANT: Precision over volume

Only flag real defects. Do not pad the review with style nits, speculation, or restatements of what the diff does. Code that looks unusual but is correct and documented is not a finding. A wrong finding is worse than a missing one.

Steps

Step 1: Read the diff

Read the diff file the user pointed at. Identify which files and functions it touches.

Step 2: Read the current source

Read the current version of every file the diff touches, so you understand the context the change lands in.

Step 3: Analyze the change

Mentally apply the diff and look for defects:

  • Off-by-one errors in loops, ranges, slicing
  • Wrong comparison operators (> vs >=, boundary conditions vs the documented behavior)
  • Logic that contradicts the function's own docstring or naming
  • Broken callers, changed return types, missed edge cases (empty input, zero, None)

Check claimed behavior against actual behavior: if the docstring says "threshold or more", the code must use >=.

Step 4: Deliver the review

Reply with this exact structure:

## Review: <diff file name>

### Findings

- <file>:<approx line>: [HIGH|MEDIUM|LOW] <what is wrong and why>
- ...or "No findings."

VERDICT: APPROVE

or, if there is at least one HIGH or MEDIUM finding:

VERDICT: REQUEST_CHANGES

The reply MUST end with exactly one VERDICT: line containing either APPROVE or REQUEST_CHANGES.

Definition of done

All of these must be true:

  • The diff file and every touched source file were read
  • Each finding names the file, approximate location, severity, and the concrete defect
  • No source file was modified and no new file was created
  • The reply ends with exactly one VERDICT: line

Error handling

  • If the diff file does not exist, say which file was expected and stop
  • If the diff does not apply cleanly to the current source (context mismatch), report that as a HIGH finding
  • If the diff is empty, reply with "No findings." and VERDICT: APPROVE

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.