design-award-match

Files (33)

• README.md
• README_EN.md
• SKILL.md
• openai.yaml · agents/
• match-input.example.json · examples/
• project-profile.example.json · examples/
• award-profile-guide.md · references/
• award-profile-schema.json · references/
• core77.json · references/awards/
• dia.json · references/awards/
• epda.json · references/awards/
• good-design-japan.json · references/awards/
• idea.json · references/awards/
• if-design.json · references/awards/
• if-student.json · references/awards/
• index.json · references/awards/
• james-dyson.json · references/awards/
• k-design.json · references/awards/
• red-dot-design-concept.json · references/awards/
• red-dot-product.json · references/awards/
• category-crosswalk.json · references/
• criteria-crosswalk.json · references/
• evidence-policy.md · references/
• matching-framework.md · references/
• output-template.md · references/
• award_profiles.py · scripts/
• build_candidate_set.py · scripts/
• filter_eligible_awards.py · scripts/
• score_award_matches.py · scripts/
• validate_award_profiles.py · scripts/
• test_award_profiles.py · tests/
• test_eligibility_and_routing.py · tests/
• test_score_award_matches.py · tests/

SKILL.md (first lines)

# Design Award Match

## Purpose

Identify the most defensible award, program, track, and category for a design project. Treat fit scores as transparent decision aids, never as probabilities of winning.

## Scope

- Extract decision-relevant project facts.
- Pre-filter the configured award allowlist with stable eligibility gates.
- Verify all dynamic requirements on current official pages.
- Compare project evidence with published criteria and observable winner trends.
- Score, rank, and explain strategic fit.
- Stop after recommendation; do not audit every submission file or redesign the project.

Use `$design-award-search` for verified same-category winners. Route file format, size, naming, declarations, and upload completeness to `$design-submission-check`.

## Input Contract

Accept a brief, images, PDF, portfolio page, or structured JSON. Extract or request only facts that can change the recommendation:

- primary function, target user, and use context;
- innovation and supporting evidence;
- project state and completion, launch, or release date;
- applicant type, student status, country or region;
- candidate awards, intended cycle, budget, and geographic constraints.

Use the canonical values in [references/category-crosswalk.json](references/category-crosswalk.json). If the primary function is unclear, ask one short question. If an eligibility fact is missing, continue with `Eligibility: Unknown`; never assume a pass.

Offer this template when the user asks how to use the skill:

```text
Project: {name and one-sentence description}
Primary function: {job performed or problem solved}
Target user / context: {optional}
Innovation and evidence: {optional}
Development status / launch date: {optional}
Applicant: {student, individual, studio, company; country/region}
Candidate awards: {optional; omit to search the supported allowlist}
Submission cycle / constraints: {optional year, region, budget}
```

## Workflow

### 1. Build the project profile

Read [../design-judge-shared/category-taxonomy.md](../design-judge-shared/category-taxonomy.md) and [references/category-crosswalk.json](references/category-crosswalk.json). Classify by primary function before appearance. Record one canonical category, no more than two adjacent categories, project state, applicant type, evidence, timing, and constraints. Label material inferences.

For command-line pre-filtering, prepare JSON like [examples/project-profile.example.json](examples/project-profile.example.json).

### 2. Validate and load the supported award set

Read [references/awards/index.json](references/awards/index.json) and [references/award-profile-guide.md](references/award-profile-guide.md). Analyze only the programs in the allowlist. Treat Red Dot Product and Red Dot Design Concept as separate programs. If a requested award is absent, return `Unsupported` rather than researching and silently adding it.

When a shell is available, validate configuration before analysis:

```powershell
python scripts/validate_award_profiles.py --pretty
```

Build a focused candidate set, normally three to five routes:

```powershell
python scripts/build_candidate_set.py examples/project-profile.example.json --limit 5 --pretty
```

Use `--award` repeatedly to restrict candidates. Award ids and declared aliases are accepted.

### 3. Apply stable gates

Use each selected profile's `routes`, `required_project_fields`, and `stable_constraints`. When needed, run:

```powershell
python scripts/filter_eligible_awards.py examples/project-profile.example.json --include-ineligible --pretty
```

Assign:

- `Eligible`: all stable and current official gates pass.
- `Ineligible`: a confirmed rule excludes the entry.
- `Unknown`: a project fact is missing or any live gate remains unchecked.

Exclude `Ineligible` routes from ranking but state the exact reason. Keep `Unknown` routes conditional.

### 4. Verify dynamic rules live

Read only the selected award profiles under `references/awards/`. For every `dynamic_gate` and relevant `dynamic_field`, verify current official pages at request time:

- cycle status and absolute deadlines;
- applicant, geography, age, enrollment, and graduation rules;
- completion, publication, distribution, or launch windows;
- exact track and category labels;
- current judging criteria;
- enough material and physical-delivery requirements to assess feasibility;
- fees and mandatory winner obligations when they affect priority.

Record direct URL and `checked on: YYYY-MM-DD`. Profile category hints are routing aids only; current official labels control. Never rely on stored dates, fees, category numbers, or remembered requirements.

### 5. Compare criteria and winner evidence

Read [references/evidence-policy.md](references/evidence-policy.md) and [references/criteria-crosswalk.json](references/criteria-crosswalk.json). Display each award's official criterion name; use normalized dimensions only for cross-award comparison.

Map every criterion to concrete project evidence using `Strong`, `Partial`, `Weak`, or `Unknown`. Do not award alignment for generic claims.

Past winners are optional evidence. Use `$design-award-search` or a small verified official-source sample. State sample size, years, category, and limitations. Describe observable `past-winner trends`, never hidden jury preferences.

### 6. Score and rank

Read [references/matching-framework.md](references/matching-framework.md). Score five dimensions from 0 to 5 with one evidence sentence per rating. Prepare input using [examples/match-input.example.json](examples/match-input.example.json), then run:

```powershell
python scripts/score_award_matches.py examples/match-input.example.json --pretty
```

Keep separate:

- `Fit score`: weighted strategic compatibility, 0–100.
- `Evidence confidence`: High, Medium, or Low.
- `Eligibility`: Eligible, Unknown, or Ineligible.

From SeanJ1ang/design-judge-skills, Apache-2.0.