review-pass
Review a diff, branch, or PR for correctness, security, and quality issues. Use when asked to review code changes, review a PR, or do a code review — especially useful in agents without a built-in review command.
- License 许可
- MIT
- 最近更新
- 2026-08-05
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- review-pass
- description
- Review a diff, branch, or PR for correctness, security, and quality issues. Use when asked to review code changes, review a PR, or do a code review — especially useful in agents without a built-in review command.
03 SKILL.md 正文
Review Pass
A portable, checklist-driven review for any diff. Produces ranked findings with concrete evidence, not vague impressions.
1. Determine scope
Figure out what's being reviewed: uncommitted working changes (git status, git diff), a branch against its base (git diff <base>...HEAD, git log <base>..HEAD), or a specific PR/file set the user named. Confirm the scope before reading further if it's ambiguous.
Complete when: the exact diff or file set under review is known.
2. Read with context
Read each changed file with enough surrounding code to understand callers, types, and existing patterns — not just the diff hunks. A change that looks wrong in isolation is often correct given context two lines outside the hunk, and vice versa.
Complete when: every changed file has been read with its relevant surrounding context, not just the patch lines.
3. Apply the checklist
Evaluate the change against four categories:
- Correctness — logic errors, off-by-one/boundary bugs, unhandled edge cases, wrong assumptions about types or state, race conditions.
- Security — injection (SQL, command, XSS), secrets committed in code, missing auth/authz checks, unsafe deserialization, path traversal.
- Simplification — dead code, needless abstraction, duplicated logic that should reuse an existing helper, over-engineering beyond what the change needs.
- Test coverage — new behavior with no test, a weakened or deleted assertion, an edge case the diff clearly introduces but doesn't cover.
Only flag something you can point to concretely — a suspicion without a mechanism isn't a finding yet.
Complete when: the diff has been evaluated against all four categories and every candidate issue has a specific file/line.
4. Verify before reporting
For each candidate finding, re-check it against the actual code: confirm the failure scenario is real (concrete input/state → wrong output/crash), not hypothetical. Drop anything that doesn't survive this check. This step exists because a plausible-sounding finding is often wrong once you trace the actual control flow.
Complete when: every remaining finding has a verified, concrete failure scenario.
5. Report
Report findings ranked most-severe-first. For each: file:line, a one-sentence statement of the defect, and the concrete failure scenario that triggers it. If nothing survives verification, say so plainly — an empty result is a valid outcome, not a failure to find something.
Complete when: findings (or their absence) have been reported with evidence for each.
04 同仓库更多 Skills
yeetslop
Remove AI-generated code slop introduced in the current branch. User-invoked only.
release-notes
Draft release notes or changelog entries from a commit or tag range. Use when asked to write release notes, draft a changelog, or summarize what changed between two versions or tags.
pr-writer
Draft a pull request title and description from the current branch's diff and commit history. Use when asked to write a PR description, draft a PR, or prepare a branch for review.
objfw
Opinionated guide for writing, debugging, reviewing, and testing Objective-C when the target uses ObjFW. Use for ObjFW classes and APIs, ObjFW-style types, ARC, exceptions, strings, collections, streams, sockets, run loops, portability, objfw-compile, or ObjFWTest in an ObjFW project.
noteday
Log today's work — a summary of the day plus any tickets worked on — as an entry in the personal Notion "Daily Log" database. Use when the user runs /noteday or asks to note down their day.
learn
Extract non-obvious learnings from the current session into AGENTS.md files, placed as close to the relevant code as possible. Use when asked to capture learnings, save session discoveries, or update AGENTS.md with what was learned.