SamyakJhaveri Multi-Agent Code Review
Structured code review using parallel subagents for thorough coverage.
Arguments
$ARGUMENTS— optional: file paths or commit range to review (default: staged changes)
Workflow
Phase 1: Identify Scope
- If arguments specify files/commits, use those
- Otherwise, review staged changes (
git diff --cached) - If nothing is staged, review unstaged changes (
git diff)
Phase 2: Parallel Review
Launch 4 subagents simultaneously, each reviewing the same changes:
- Style Agent — naming, formatting, consistency with existing code
- Correctness Agent — logic errors, off-by-one, null/undefined risks
- Security Agent — injection, path traversal, secret exposure, OWASP top 10
- Performance Agent — unnecessary allocations, O(n²) where O(n) is possible
Phase 3: Synthesize
- Collect all findings
- Deduplicate overlapping concerns
- Prioritize by severity: critical > high > medium > low
Phase 4: Report
Present findings grouped by severity:
## Critical
- [issue + file:line + suggestion]
## High
- ...
## Medium / Low
- ...
## Summary
N issues found (X critical, Y high, Z medium/low)
Recommendation: APPROVE / NEEDS CHANGES