/review
Run a multi-agent code review on recent changes or specified files.
Usage
code
/review # Review uncommitted changes /review <file/dir> # Review specific files or directory /review --full # Full project architecture review
Workflow
- •
Identify scope: Determine what to review:
- •No args:
git diffandgit diff --cachedfor uncommitted changes - •File/dir arg: the specified paths
- •
--full: entire project structure
- •No args:
- •
Select reviewers: Based on what changed, spin up relevant review agents as a team:
- •
.gofiles (handlers/services/database) →code-reviewer - •
.gofiles (handlers) + templates →htmx-debugger - •Auth, middleware, secrets →
security-reviewer - •Docker, Makefile, CI →
devops-reviewer - •Cross-cutting changes →
systems-architect - •Template/CSS/UX changes →
ux-reviewer - •SQL, queries, performance-sensitive code →
performance-analyst - •ML/LLM/embedding code →
ml-engineer
- •
- •
Run review team: Launch selected agents in parallel using TeamCreate. Each agent reviews the changed files through their specialist lens.
- •
Compile report: Collect findings from all agents and produce a unified, deduplicated report:
- •Critical — Must fix before merge
- •Important — Should fix, creates tech debt if deferred
- •Suggestion — Nice to have, style/optimization
- •
Create issues: For any Critical or Important findings, create beads issues to track the fixes (
bd create --title="[Bug] ..." --type=bug --priority=1for Critical,--priority=2for Important). Do not just report findings without tracking them.