Pick 10 Diverse

Input: $ARGUMENTS

Core Principles

1. Diversity is coverage, not randomness. Random selection sometimes produces diverse results, sometimes clusters. Diverse selection guarantees coverage by construction — it is an anti-clustering algorithm.

2. Five dimensions of diversity. Skills vary by category, tier, function (analyze/decide/plan/create/validate), workflow phase (early/mid/late), and connectivity (standalone vs hub). True diversity covers all five dimensions, not just category.

3. Round-robin prevents dominance. If you pick the “best” from each category, large categories dominate because they have higher-quality peaks. Round-robin across categories, then within categories by tier, prevents this.

4. Mandatory function coverage. The 10 picks must include at least one skill from each of these functions: analysis, decision, planning, validation, and exploration. Missing any function creates a blind spot in the user’s toolkit.

5. Surprise has learning value. A diverse set should include at least one skill the user probably hasn’t encountered. If all 10 are well-known, the diversity is only surface-level.

Phase 1: Dimension Mapping

[A] CATEGORY_MAP:

Step 1: List all categories in skills.json with skill counts
    | Category | Skill Count | Representative Skills |
    |----------|-------------|----------------------|
    | [cat]    | [N]         | [top 3 by tier]      |

Step 2: Identify underrepresented categories (fewer than 5 skills)
Step 3: Note: underrepresented categories get priority — their skills are rarer

[B] FUNCTION_MAP:
    ANALYSIS: [skill IDs whose primary purpose is understanding/decomposing]
    DECISION: [skill IDs whose primary purpose is choosing/comparing]
    PLANNING: [skill IDs whose primary purpose is sequencing/scoping]
    CREATION: [skill IDs whose primary purpose is producing/writing]
    VALIDATION: [skill IDs whose primary purpose is checking/testing]
    EXPLORATION: [skill IDs whose primary purpose is discovering/generating]
    DIAGNOSIS: [skill IDs whose primary purpose is finding problems/causes]

[C] TIER_MAP:
    tier1 (core): [count]
    tier2 (important): [count]
    tier3 (domain): [count]
    tier4 (specialized): [count]
    category (router): [count]
    experimental: [count]

Phase 2: Round-Robin Selection

[D] SELECTION:

Step 1: Order categories by size (smallest first — rare categories get first pick)
Step 2: Round-robin: take the highest-tier skill from each category
    - Skip categories already represented
    - Continue until 10 slots filled or all categories visited
Step 3: If fewer than 10 categories: take second-best from largest categories
Step 4: If more than 10 categories: prioritize categories covering unfilled functions

[E] FUNCTION_COVERAGE_CHECK:
    After selection, verify:
    - Analysis represented? [Y/N — if N, swap weakest pick for best analysis skill]
    - Decision represented? [Y/N — if N, swap]
    - Planning represented? [Y/N — if N, swap]
    - Validation represented? [Y/N — if N, swap]
    - Exploration represented? [Y/N — if N, swap]

Phase 3: Diversity Verification

[F] DIVERSITY_METRICS:

    Categories covered: [N] out of [total]
    Tiers covered: [list]
    Functions covered: [N] out of 7
    Phases covered: [early/mid/late — which are present]
    Connectivity mix: [N standalone] + [M hub skills]

    SURPRISE_CHECK:
    - How many of these 10 are well-known (tier1/tier2 with high connectivity)? [N]
    - If N > 7: swap 1-2 well-known for lesser-known skills from underrepresented areas

[G] DIVERSITY_SCORE:
    = (categories_covered / total_categories × 25)
    + (functions_covered / 7 × 25)
    + (tiers_covered / 6 × 25)
    + (surprise_picks / 10 × 25)
    = [0-100]
    Target: >= 70

Phase 4: Output

ALGORITHM: DIVERSE
POOL: [total skills in library]
PICKED: 10
DIVERSITY SCORE: [X/100]

  1. /[id] — [title]
     Category: [cat] | Tier: [tier] | Function: [func]
     [1-line description]

  2. /[id] — [title]
     Category: [cat] | Tier: [tier] | Function: [func]
     [1-line description]

  [continue to 10...]

COVERAGE:
  Categories: [N/total] — [list covered]
  Functions: [N/7] — [list covered]
  Tiers: [list covered]
  Phases: [list covered]

GAPS:
  Categories NOT covered: [list]
  Functions NOT covered: [list — if any]
  To fill gaps, add: /[suggested skill for each gap]

SUGGESTED EXPLORATION ORDER:
  Start with: /[most accessible skill]
  Then try: /[skill that builds on first]
  Deep dive: /[most specialized pick]

Failure Modes

Depth Scaling

Default: 2x. These are floors.

Pre-Completion Checklist

• All categories in library identified and counted
• Round-robin selection performed (smallest categories first)
• All 5 mandatory functions represented (analysis, decision, planning, validation, exploration)
• Exactly 10 skills returned
• At least 2 lesser-known or surprise picks included
• Diversity score calculated and >= 70
• Gaps explicitly identified with suggested additions
• Exploration order suggested

Integration

• Shortcut for: /pick 10 diverse
• Use when: You want a broad sample of the skill library, not a targeted search
• Routes to: The 10 picked skills; /p10useful for quality-ranked alternative
• Related: /p10random (random, not guaranteed diverse), /p10useful (quality-ranked)
• Differs from /p10random: random may cluster; diverse guarantees coverage
• Differs from /p10useful: useful ranks by quality; diverse ranks by coverage
• Differs from /p10complement: complement centers on one skill; diverse covers the whole library