Agent
Frontend-only browser QA tier — drive a real browser to verify a web change actually renders, responds to events, and makes the right network calls. Use after a web-frontend change (alongside code review, in parallel) to confirm the UI works in a browser, not just that tests pass. Prefers agent-browser, falls back to…
Agent
Implement an approved flow plan — work tasks.md top to bottom with a TDD cycle and atomic Conventional Commits, leaving the branch ready for deploy. Use when the plan and task list exist and the user has signed off, and you want the implementation run in a cost-efficient subagent instead of the frontier orchestrator.
Agent
Turn one framed flow task into a lightweight, visual implementation plan — plan.md (Mermaid-led) plus a tasks.md checklist — before any code is written. Use when you want planning delegated to a fresh frontier-model context instead of running it in the orchestrator. Reads brief.md / research.md if present. Plans; does…
Agent
Frontend-only React/Next.js quality reviewer — audits component design for composition and reuse, not just correctness. Use on React changes (alongside code review, in parallel) to catch boolean-prop proliferation, one-off components that should be composed/reused, and design tokens defined top-down instead of from…
Agent
Cost-efficient pre-plan investigation subagent — codebase exploration, GitHub issue/PR/commit-history search, or web/library lookups. Fan several out in parallel, one per area, so the frontier orchestrator does not crawl the repo itself. Returns a tight decision-relevant digest, never raw dumps. Read-only.
Agent
In-harness fresh-eyes reviewer of a plan and its implementation (or a diff / PR). Use for a fast local review pass when you do not want to spin up external agents — it complements, and does not replace, the code-review-brief → user-run-external-agents path. Read-only; reports findings, does not fix or post.