// Orchestrate a multi-round implementation + review cycle. Use when coordinating a feature that requires implementation (FE/BE), design review (UIUX), security review, and QA verification. Ensures QA findings get routed back for fixes until clean.
Connect to Chrome via CDP proxy to automate web interactions — posting, scraping, form filling. Uses puppeteer-core (no bundled Chromium).
Short description — what this does and when to use it.
Review recent edits and update all documentation including architecture docs, API specs, and edit history. Creates missing docs for new implementations.
Review recent edits and update all documentation including architecture docs, API specs, and edit history. Creates missing docs for new implementations.
Review code for best practices, modularity, scalability, abstraction, test coverage, redundancy, hardcoded values, type safety, performance, naming, API design, async patterns, config/env sync, template consistency, and documentation alignment. Generates detailed report with issues and recommendations.
| name | review-loop |
| description | Orchestrate a multi-round implementation + review cycle. Use when coordinating a feature that requires implementation (FE/BE), design review (UIUX), security review, and QA verification. Ensures QA findings get routed back for fixes until clean. |
Orchestrate implementation through multiple rounds until QA reports zero issues. This prevents the one-shot delegation problem where QA finds bugs but nobody fixes them.
Use this when you're a coordinator (Dev Lead, PM) assigning a feature that involves multiple specialists.
Identify all stakeholders — before delegating, ask: who needs to be involved?
Delegate design review first (if UI work):
delegate_task_async → UIUX Designer: "Review the interaction design for [feature]"
Delegate implementation (after design, or parallel if non-UI):
delegate_task_async → Frontend Engineer: "Implement [feature] following UIUX spec"
delegate_task_async → Backend Engineer: "Add [endpoint]" (if needed)
delegate_task_async → Security Auditor: "Review [feature] for [specific concerns]"
Delegate QA (can start in parallel — QA reads existing code while FE works):
delegate_task_async → QA Engineer: "Review [feature], run full test suite, write missing tests, grep for convention violations"
Collect all results via check_task_status on each delegation.
If QA reported issues:
Send QA's findings back to the implementer:
delegate_task → Frontend Engineer: "QA found these issues in your implementation:
[paste QA's specific findings with file:line references]
Fix all of them and report back."
Re-run QA on the fixes:
delegate_task → QA Engineer: "FE applied fixes for your findings. Re-verify:
[paste the specific issues that were fixed]
Run the test suite again. Report if any issues remain."
If QA still finds issues → repeat Round 2.
When QA reports zero issues:
delegate_task_async to all specialists at once, then collect with check_task_status. Don't serialize what can be concurrent.delegate_task directly to BE — don't relay through the lead.