orchestrate

Files (1)

• SKILL.md

SKILL.md (first lines)

# Orchestrate

**Hard gates — never skip:** (1) Developer always hands off to quality. (2) Quality must produce qa-report.md and review.md. (3) Never declare work done without quality artifacts. (4) Never infer quality from developer output. (5) Never silently skip a workflow step.

Default to `prototype-first`: use planner-to-developer-to-quality for the smallest valuable first feature, then reach a testable prototype without routine confirmations. Use developer-to-quality for small bounded changes. Keep human updates short with status, essential risks, and next action. Before classifying, diagnose whether the problem is understood: if the user reports something broken, first establish where the problem actually is (code, config, environment, infrastructure, user error) before routing any code changes. Never modify code to fix a problem that has not been diagnosed. Classify into tiers: Tier 1 (clearly bounded, diagnosis confirms code issue) routes developer then quality. Tier 2 (probably small, scope unclear) routes planner for 1-2 questions then developer. Tier 3 (clearly complex) routes full planner workflow. Default to tier 2 when unsure. On every handoff, identify which skills match the work and tell the subagent to load them — the user should never need to request a skill explicitly. Route UI work to `frontend-design` (or `ui-ux`), test work to `unit-testing`/`feature-testing`/`e2e-testing`, review to `quality-review`, unclear scope to `requirements-gathering`, architecture to `solution-design`, tickets to `task-breakdown`, Git to `branch-safely`, deploy to `deploy` (plus `deploy-vercel` or `deploy-vps` as appropriate, or `deployment-decisions` if no process exists), docs to `documentation`, production to `monitoring`. Do not skip routing for small changes. Before developer work, relay only the planner questions needed for goal, users, constraints, smallest useful result, and success criteria; then get explicit approval of the first-feature plan. Never infer approval from silence. Ask for approval only for a material scope/goal change; security, cost, privacy, credential, destructive, remote, or deployment decision; or an explicit user-requested checkpoint. Guided/checkpointed and autonomous delivery are opt-in. Planner initializes `AGENTS.md` from the template when absent. Every implementation must include tests at all three applicable levels: unit tests for isolated logic, feature tests against acceptance criteria, and end-to-end tests simulating real user flows. For a bug fix, developer writes regression tests at each level before the fix; quality verifies all three pass. Update state, backlog state, and project memory after handoffs. At every completed user-facing feature, summarize the exact local commands and URL from `.agents/artifacts/local-test.md` organized by test level, then ask whether the user wants to test, fix, adjust, or start the next feature. Do not automatically plan or build the next major feature unless autonomous mode was explicitly selected. If quality reports `BLOCKED` or `PASS_WITH_NOTES`, state what is missing and do not present the work as fully complete. Summarize specialist evidence and retain detail in artifacts. Do not claim unproduced tests, commands, URLs, or results. Retry a failed agent once, then ask the human.

From prinx/agents, MIT.

More from prinx/agents