| name | design-award-pipeline |
| description | Route and coordinate an end-to-end design-award workflow across winner research, evidence-based evaluation, award matching, entry-text preparation, and final submission checking. Use when a user asks for a complete award plan, does not know which Design Judge skill to use, wants multiple stages coordinated, or needs a resumable workflow with explicit handoffs. Do not replace the specialist skills, invent project facts, treat scores as winning probabilities, or bypass current official-rule verification. |
Design Award Pipeline
Coordinate the specialist Design Judge skills without duplicating their rules.
Route the request
Choose the smallest sufficient route:
| User goal | Route |
|---|
| Find comparable winners | design-award-search |
| Diagnose or score the design | design-evaluation |
| Select an award, route, or category | design-award-match |
| Prepare entry fields and copy | design-information-prep |
| Audit a concrete submission package | design-submission-check |
| Complete journey | evaluation or match → information prep → submission check; add search only when precedents are needed |
If the user has not provided enough information to select a route, ask at most one short question. Otherwise proceed with explicit assumptions.
Maintain the handoff
Create or update a compact handoff record using references/handoff-schema.json. Keep these concepts separate:
- user-supplied facts;
- evidence-backed findings;
- model inferences;
- missing facts and live-rule checks;
- decisions already approved by the user;
- the next recommended specialist skill.
Do not copy long specialist outputs into the handoff. Store identifiers, decisions, blockers, source links, and artifact paths.
When a handoff is written to JSON, run python scripts/validate_handoff.py <handoff.json> before treating it as resumable. Use examples/handoff.example.json as a minimal valid shape, not as project evidence.
Coordinate stages
- State the selected route and why it is sufficient.
- Invoke or follow the relevant specialist Skill exactly.
- Preserve its uncertainty labels, evidence confidence, and blockers.
- Stop for user approval when the target award, maturity track, or another consequential choice changes downstream work.
- Hand off only facts supported by user material or cited official sources.
- End with the completed stage, unresolved blockers, and the next optional stage.
Boundaries
- Never merge fit score, design score, and evidence confidence into one number.
- Never claim or estimate a probability of winning.
- Never treat observed winners as official jury weights.
- Never mark a package ready when time-sensitive official rules remain unverified.
- Never invoke all specialist skills when one is enough.
