| name | plan |
| description | Plan orchestrator that classifies coding tasks and dispatches the best skills in .agents/skills for features, refactors, bug fixes, tests, docs, and delivery. |
| argument-hint | Task goal + constraints, e.g. 'add workspace sharing with tests' or 'refactor DAL and keep API stable' |
Plan - Skill Router
Classify request, choose minimal skill set, and execute in phases.
Route Categories
| Category | Trigger patterns | Primary skills |
|---|
| New feature | add/create/build/implement/new flow/resource | cook, physthink-feature, frontend-development, backend-development |
| Refactor | refactor/restructure/cleanup/tech debt/no behavior change | frontend-development, backend-development, code-review, web-testing |
| Bug fixing | bug/error/failing/flaky/broken/regression | bug-fix, debugging, web-testing |
| Testing | add tests/coverage/e2e/unit/integration/perf | web-testing, code-review |
| UI/UX polish | redesign/layout/style/accessibility/motion | frontend-design, ui-styling, aesthetic, shadcn |
| Data/API work | prisma/schema/migration/query/router/auth | physthink-feature, databases, backend-development, better-auth |
| Research/docs first | unfamiliar lib/api/standard/architecture | docs-seeker, sequential-thinking |
Dispatch Rules
- Detect intent from verbs and constraints.
- Select
1-4 skills only; avoid over-selection.
- Prefer existing orchestrators first:
- Feature-heavy request -> start with
cook.
- Defect-focused request -> start with
bug-fix.
- Add specialist skills only for explicit scope (auth, db, UI, perf, 3D, payments).
- If request is ambiguous, run
sequential-thinking before implementation.
- Before coding, present the short plan and wait for approval.
Execution Phases
- Scope
- Extract outcome, constraints, non-goals, acceptance checks.
- If missing success criteria, define measurable checks before coding.
- Plan
- Build a short step plan across layers (data -> API -> UI -> tests).
- Identify risks and required migrations/config updates.
- Request approval before starting implementation.
- Implement
- Execute selected skills in dependency order.
- Keep edits minimal and consistent with local patterns.
- Verify
- Run targeted tests/lint/typecheck for changed areas.
- If no tests exist, add at least one regression or happy-path test.
- Review
- Apply
code-review mindset before claiming completion:
- List defects/risks first, then summary.
Skill Selection Matrix
| Need | Add skill |
|---|
| Next.js App Router/RSC | web-frameworks |
| New Prisma model or oRPC chain | physthink-feature |
| Auth/session/OAuth/RBAC | better-auth |
| Query/index/perf/migrations | databases |
| Component composition/theming | shadcn, ui-styling |
| Design direction and interaction quality | frontend-design, aesthetic |
| CI/CD, deploy, containers | devops |
| Stuck or high uncertainty | problem-solving, sequential-thinking |
Completion Gate
Do not mark complete until all are true:
- Requested behavior works end-to-end.
- Relevant tests pass or explicit gap is documented.
- Potential regressions and risks are stated.
- Files changed and why are summarized concisely.
Example Routes
| Request | Route |
|---|
| "Create course draft publishing flow" | cook -> physthink-feature -> frontend-development -> web-testing |
| "Refactor DAL without API break" | backend-development -> code-review -> web-testing |
| "Fix login redirect loop" | bug-fix -> better-auth -> web-testing |
| "Improve dashboard visual hierarchy" | frontend-design -> ui-styling -> aesthetic |
| "Prisma query is slow" | bug-fix -> databases -> web-testing |
Load Rule
Read selected skill instructions using this path order:
.agents/skills/<skill>/SKILL.md~/.copilot/skills/<skill>/SKILL.md~/.claude/skills/<skill>/SKILL.md
If the skill exists in multiple places, prefer workspace-local (.agents) version.
Load only additional references/scripts that are needed for the current step.
