← ClaudeAtlas

pr-reviewlisted

Multi-lane review of a single PR or diff for any backend project — runs real automated security/dependency/secret scans, launches the project's reviewer agent and the audit skill's security checks, then a quality pass, and combines everything into a severity-ranked report with an explicit blocking verdict. Scoped to the diff (per-PR gate); for a full-codebase safety net use /audit. Treats the codebase as sensitive by default (fintech-grade).
YoussefRashad/backend-agent-skills · ★ 0 · Code & Development · score 72
Install: claude install-skill YoussefRashad/backend-agent-skills
# PR Review — Multi-Lane Diff Review & Vulnerability Scan > **When to use:** gating a single PR or diff before merge. Invoke with `/pr-review <PR-url | #number | branch | paths>` (no argument = current branch vs base), or "review this PR". For the whole codebase instead of a diff, use `/audit`. Run three independent lanes on the **same diff**, triage the results, and produce one severity-ranked report with a blocking verdict. Automated lanes catch what eyes miss; human-style lanes catch what tools miss. None alone is sufficient. ## Critical Rules (read first) 1. **Never read or review secret files:** `.env`, `.env.*`, `**/.env*`, `*.pem`, `*.key`, `*.crt`, `*.cer`, `*.p12`, `*.pfx`, `*.jks`, `*.keystore`, `**/id_rsa*`, `**/secrets/**`, `**/credentials/**`, `**/.aws/**`, `**/.ssh/**`, `**/.gnupg/**`. If a secret appears **inside the diff**, that is itself a Critical finding — report the file:line and the secret _type_, never echo the value. 2. **Review only what changed.** Scope to the diff. Do not review `node_modules/`, `dist/`, `build/`, generated files, lockfiles (except to flag dependency risks), or migration snapshots — unless the change is _in_ them. 3. **Every scanner finding is a candidate, not a verdict.** Triage for false positives before reporting (see Step 4). Label confidence. 4. **Never auto-fix or commit.** Review only. Suggest fixes as diffs/instructions. 5. **Fail loud, not silent.** If a lane can't run (tool missing, skill absent), say so in the report —