← ClaudeAtlas

code-review-checklistlisted

Principal-level code review procedure across correctness, error handling, concurrency, security, API design, performance, tests and readability, with a severity rubric, a file:line + failure-scenario + fix finding format, and a verify-before-report protocol. Load when reviewing a diff, PR, branch range or module, or when triaging review findings produced by others.
rustcircle/c101 · ★ 0 · Code & Development · score 65
Install: claude install-skill rustcircle/c101
# Code review checklist A review exists to find defects that would reach users, and to say so in a form the author can act on in one pass. Every finding you report is a claim with a concrete failure scenario and evidence. A suspicion without a scenario is a question, not a finding. ## 1. Scope the review before reading code 1. Identify the target and its size. ```bash git diff --stat origin/main...HEAD # three dots = changes since merge-base git log --oneline origin/main..HEAD gh pr view 123 --json title,body,files,baseRefName && gh pr diff 123 git diff -U10 origin/main...HEAD -- src/ # extra context lines around hunks ``` 2. Write one sentence: "This change intends to <X> by <Y>." If you cannot, read the PR body, linked issue and commit messages; if still unclear, that is an Open doubt for the caller. 3. Budget by size of *logic* changed (exclude lockfiles, generated code, snapshots, vendored code). | Logic lines changed | Approach | |---|---| | < 400 | One pass, all dimensions, by you | | 400 - 1500 | Split by subsystem; review each chunk fully before the next | | > 1500, or > 30 files | Main thread only (no `<stack-context>` header): `review-verify` workflow with `target` + explicit `dimensions`. As a subagent: shard per subsystem and spawn one `reviewer` per shard as parallel Agent calls in one message with `run_in_background: false`, or return `Workflow name=review-verify args={"target":"<range>","dimensions":[...]}` as