review-diff
Skill1 starsChanged 44 days agoMITView on GitHub
Reviews the current diff against the repo's quality preset, reporting findings as file:line with a severity. Use to self-review before a PR, or to review uncommitted or staged changes.
19 other projects use a version of this.
Adopted by
Most recently taken up by
- jisuzzzz/haechitheir own version, first seen Nov 8, 2026
- anorth/agent-skillstheir own version, first seen Nov 3, 2026
- Hy0sh/claude-skillstheir own version, first seen Nov 3, 2026
- tounsils/claude-code-skillstheir own version, first seen Oct 25, 2026
- nesnilnehc/ai-cortextheir own version, first seen Oct 25, 2026
- bbvch-ai/aihub-coretheir own version, first seen Oct 25, 2026
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.
GoogleSkillgrilling
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
Other versions19 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 19
- faizkhairi/claude-code-blueprint★ 73same name
- huytd/skills★ 19same name
- jisuzzzz/haechi★ 17same name
- bbvch-ai/aihub-core★ 16same name
- nesnilnehc/ai-cortex★ 8same name
- tardigrde/agent-skill-eval★ 4same name
- Hy0sh/claude-skills★ 2same name
- Pradnyeshp/Claude-Skills★ 2same name
- anorth/agent-skills★ 1same name
- erikediner/skills★ 1same name
9 more not listed
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 repository9 other files it gives its agents.
- address-reviewSkill
- draft-prSkill
- frame-productSkill
- implement-changeSkill
- plan-changeSkill
- trace-incidentSkill
- triage-taskSkill
- write-commitSkill
- write-handoverSkill
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 0 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 19 edited versions we read keep these.
0
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
- Very specific: line 30
Check the rest of the gates yourself: Every gate review context printed that review gates did not settle is a question with a yes or no …
- Very specific: line 62
Every finding cites
file:lineand quotes the line: Read the line to quote it.
“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.
Files
review-diff
wb <args> # wb not on PATH: python <this skill's directory>/../../lib/wb.py <args>Steps
- Get the computed facts.
review context(add--stagedfor the staged diff). It reports the preset, every changed file, the critical zones touched, the sources that changed with no test alongside, and the gates that apply. - Read the diff itself, then read the surrounding code. A diff shows what
changed, not what it broke. For anything that changes a shared contract, ask
the index who calls it:
../../shared/code-search.md. - Settle the mechanical gates first.
review gates(add--key <KEY>so the bug-fix regression rule knows the ticket type). It reads the added lines and reports secrets, swallowed errors and added dependencies asfile:linefindings. Exit 7 means it found some; report them, do not re-derive them. A dependency finding is a question to answer, not a defect: say who owns it and why nothing already in the repo does the job. Check the rest of the gates yourself. Every gate
review contextprinted thatreview gatesdid not settle is a question with a yes or no answer for this diff. The floor is not negotiable at any preset: unit test for changed logic, regression test with a bug fix, no swallowed error, no secret, a stated rollback.review contextalso lists files the plan gained after its audit (plan amended after its audit). The first audit never saw those files, so read them as closely as a critical zone.- Go deeper in critical zones. Billing, auth, user data, migrations and secrets are held to a higher standard than the preset otherwise sets. In those files, read every changed line and its error path.
Write the findings, most severe first, to
.workflow/<KEY>/review.md:- high `src/checkout.py:42` — `charge = create_charge(total)` Coupon is validated after the charge. An expired coupon charges, then errors.Severity: high — wrong behaviour, data loss, or a security hole. medium — will break under a plausible input or state. low — real but contained. Skip formatting and taste unless it changes meaning. Then
cite check .workflow/<KEY>/review.md --worktree, and report only once it passes: a finding whose line is not there is withdrawn, not reworded.- Say when it is clean. "No findings" is a result. Do not manufacture a finding to look thorough.
Rules
- Every finding cites
file:lineand quotes the line. Read the line to quote it; do not quote from an index or from memory. - State the failure, not the preference. A finding needs concrete inputs or state that produce a wrong outcome. "Consider extracting this" is not a finding.
- The missing-test list is a heuristic, not a verdict. Check each one: a test may cover it under another name.
- Do not fix while reviewing. Report; let the author decide.
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
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.