Semgrep + CodeRabbit Review

Two-stage code review: fast deterministic pattern detection first, then semantic AI analysis. Order is non-negotiable.

NEVER

Never run CodeRabbit before Semgrep passes — Semgrep failures are blockers, not suggestions.
Never skip re-running both tools after fixes — fixes can introduce new issues.
Never treat LOW findings as blocking — they are optional polish.
Never do more than 3 review cycles on the same PR — if still failing after 3, break the PR into smaller pieces.
Never context-switch between issue types mid-fix — batch similar issues together.

Execution Order

Stage 1: Semgrep (10-20 seconds)
  └─ FAIL → fix ALL violations → re-run until PASS
  └─ PASS → proceed to Stage 2

Stage 2: CodeRabbit (5-30 minutes)
  └─ Evaluate findings by priority
  └─ Fix CRITICAL + HIGH before merge
  └─ Consider MEDIUM if reasonable
  └─ LOW is optional

Stage 3: Verify
  └─ Re-run both tools after fixes
  └─ Confirm no new issues introduced

Commands

# Stage 1 — run on each changed file (loop required for Semgrep 1.146.0+)
for f in $(git diff --name-only HEAD); do
  semgrep scan --config auto --json "$f" 2>/dev/null
done

# Stage 2
coderabbit review --diff uncommitted

# Verify after fixes (repeat both)
semgrep scan --config auto <changed-files>
coderabbit review --diff uncommitted

If a project has a .semgrep.yaml, use --config .semgrep.yaml instead of --config auto.

Priority Reference

Priority	Level	Examples	Deadline
CRITICAL	Blocking	Hardcoded secrets, auth bypass, SQL injection, XSS, data leakage	Must fix before merge
HIGH	Architectural	Missing auth guards, schema mismatch, breaking changes, race conditions	Fix before merge
MEDIUM	Quality	Weak crypto, poor error handling, type safety violations, duplication	Fix if reasonable
LOW	Polish	Suggestions, optimization opportunities, style	Consider optional

What Each Tool Catches (non-obvious)

Semgrep: Hardcoded secrets/API keys, missing auth guards on routes, weak crypto (MD5/SHA1), debug statements in prod, unsafe any types, SQL injection patterns.

CodeRabbit: N+1 query patterns, multi-tenant isolation gaps, API contract drift between services, test coverage gaps on behavioral changes — things that require semantic understanding of the codebase.

Fix Strategy

Batch by type, not by file. Fix all secrets issues, then all auth issues, then all injection issues. Run unit tests after each batch.

CRITICAL fixes first — security and auth bypass issues must be resolved before anything else. They are non-negotiable blockers.

3-pass rule — if the review-fix-reverify cycle exceeds 3 iterations, the PR scope is too large. Split it.

Conflict Resolution

When Semgrep and CodeRabbit give conflicting feedback on the same line: trust Semgrep. Its rules are deterministic and explicit. Validate CodeRabbit's concern with context before acting on it.

When to Use Each Mode

Situation	Use
Pre-commit quick check	`--semgrep-only` (10-20s)
Before creating PR	(empty) full two-stage
Review specific recent commits	`--since HEAD~3`
Only staged files (partial work)	`--staged`
CodeRabbit not installed/available	`--semgrep-only` + manual checklist

Arguments

$ARGUMENTS: Optional user-provided target, path, environment, symptom, or constraint. When empty, infer the narrowest safe scope from the current repository context and ask only if multiple high-impact choices remain.

/semgrep-coderabbit

SKILL.md