// Orchestrate complex tasks with agent teams. Clarify requirements, plan with dual-model review, implement with TeamCreate, simplify, review, and ship.
Use for complex or high-risk implementation work that needs plan-first execution, explicit verification design, pre-code review, and autonomous post-implementation hardening through self-test, simplify, and review-loop.
Prove the requested goal works on the highest-signal affected surface before closing. Default closing behavior for every implementation task — not just when explicitly asked. Also triggers on: 'make sure it works', 'test it yourself', 'don't just tell me it's done', 'I don't want to test this'.
Post-implementation review/fix loop. Use host-appropriate review coverage on your implementation, fix issues, iterate until highly confident, then commit.
Lightweight read-only audit. Use host-appropriate review coverage for evidence and findings. Iterate on disagreement or low confidence. No code changes.
Iteratively clarify requirements for a new task until reaching 90%+ confidence, then auto-invoke plan-loop.
Audit agent skills in a repo against best practices — checks .claude/skills/ and .agents/skills/ for quality, structure, and coverage gaps. Use when the user wants to review, improve, or evaluate their skills, or when they mention skill quality, skill audit, or skill best practices.
| name | orchestrator |
| description | Orchestrate complex tasks with agent teams. Clarify requirements, plan with dual-model review, implement with TeamCreate, simplify, review, and ship. |
End-to-end flow for complex tasks: clarify → plan → implement with teams → simplify → review → ship.
For simple tasks (1-3 files), skip this — just implement directly, then review-loop.
Consider $ARGUMENTS if provided. Read ~/dev/agent-guards/AGENTS.md before starting.
You drive each phase and control all transitions. When a sub-skill finishes, return here for the next phase.
Use the Skill tool to invoke new-task. When 95%+ confident, proceed to Phase 2 yourself.
Use the Skill tool to invoke plan-loop. When user approves the plan, proceed to Phase 3 yourself.
Use TeamCreate for coordinated parallel work (shared task list, peer messaging). Split agents by module/feature, not by role — each agent owns a vertical slice with exclusive file ownership. Size the team to the task (2-4 agents typical).
Use the Agent tool only for isolated read-only research/exploration, not implementation.
If new facts invalidate the plan, update the plan file before continuing.
Use the Skill tool to invoke simplify on the full changeset. Fix reuse, quality, and efficiency findings. This catches duplication and code smell that implementation agents introduce.
Use the Skill tool to invoke self-test. Run the verification plan defined during Phase 2 against the real surfaces affected by the change. If verification fails, fix and re-verify before proceeding. Do not skip this phase — code review is not a substitute for proving it works.
Use the Skill tool to invoke review-loop on the full changeset (including simplify and verify changes). At >=85 confidence: commit, push, and create PR. Include in the PR summary: what was built, key decisions, and residual risks.
