Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Maximize portfolio diversity and coverage using MAP-Elites, Niche coverage, Maximum dispersion, and Anti-clustering methods.
.claude/skills/yogsoth-ai-diversity-maximization/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -12% | 0% |
Select a portfolio that maximizes coverage across the solution space, avoiding redundancy and ensuring representation of distinct approaches, domains, or capabilities.
| Dimension | Target | |-----------|--------| | Candidates evaluated | 8-20 | | Niches defined | 4-10 | | Diversity dimensions | 2-5 | | Coverage threshold | >=80% of defined niches |
| Field | Type | Description | |-------|------|-------------| | candidates | list | All candidates with feature vectors | | niches | list | Defined niches or capability areas | | coverage_map | matrix | Which candidates cover which niches | | diversity_score | number | Aggregate diversity metric | | gaps | list | Uncovered or under-covered niches |
| Tactic | When | |--------|------| | niche-coverage-analysis | Need to map and score coverage systematically | | pareto-frontier-construction | Trading off diversity against cost or value |
| SOP | Purpose | |-----|---------| | niche-definition | Define the niches to cover | | niche-mapping | Map candidates to niches | | coverage-scoring | Score how well portfolio covers space | | objective-definition | Define diversity objectives formally |
yamlstrategy: diversity-maximization selected_portfolio: - candidate: <name> niches_covered: [<niche1>, <niche2>] coverage_score: <0-1> redundancy_score: <0-1> gaps_remaining: - niche: <name> severity: <high|medium|low> method_used: <MAP-Elites|niche-coverage|max-dispersion>
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | niche-coverage-analysis | Define niches within the solution space, map candidates to niches, score coverage completeness, and identify gaps requiring attention. | | pareto-frontier-construction | Build the Pareto frontier from multi-objective optimization, visualize trade-offs, and select a portfolio from non-dominated solutions. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | coverage-scoring | Compute coverage completeness, redundancy, and gap severity scores from a coverage map. | | niche-definition | Define niches and capability areas that a portfolio should cover based on domain structure and objectives. | | niche-mapping | Map each candidate to the niches it covers, indicating strength of coverage for each assignment. | | objective-definition | Define optimization objectives, constraints, and trade-off preferences from context and candidate information. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 20,886 | 13,270 | -36% | 1 | 1 | 0% | 3,995 | 3,595 | -10% | 0 | 0 | — |
case-02 | fail→pass | 18,423 | 18,065 | -2% | 1 | 1 | 0% | 3,477 | 4,359 | +25% | 0 | 0 | — |
case-03 | fail→pass | 24,242 | 11,254 | -54% | 1 | 1 | 0% | 4,677 | 2,961 | -37% | 0 | 0 | — |
case-04 | pass→pass | 18,738 | 12,400 | -34% | 1 | 1 | 0% | 3,815 | 3,130 | -18% | 0 | 0 | — |
case-05 | pass→fail | 18,978 | 27,526 | +45% | 1 | 1 | 0% | 4,014 | 6,173 | +54% | 0 | 0 | — |
case-06 | pass→pass | 12,506 | 22,722 | +82% | 1 | 1 | 0% | 2,584 | 5,164 | +100% | 0 | 0 | — |
case-07 | fail→pass | 11,433 | 4,126 | -64% | 1 | 1 | 0% | 1,805 | 1,485 | -18% | 0 | 0 | — |
case-08 | fail→pass | 30,967 | 1,999 | -94% | 1 | 1 | 0% | 1,240 | 1,097 | -12% | 0 | 0 | — |
case-09 | pass→pass | 11,700 | 4,399 | -62% | 1 | 1 | 0% | 1,744 | 1,515 | -13% | 0 | 0 | — |
case-10 | fail→pass | 26,900 | 1,967 | -93% | 1 | 1 | 0% | 2,087 | 1,055 | -49% | 0 | 0 | — |
case-11 | pass→pass | 7,181 | 2,445 | -66% | 1 | 1 | 0% | 1,103 | 1,163 | +5% | 0 | 0 | — |
case-12 | fail→pass | 10,979 | 2,326 | -79% | 1 | 1 | 0% | 1,736 | 1,176 | -32% | 0 | 0 | — |
case-13 | fail→pass | 14,598 | 2,429 | -83% | 1 | 1 | 0% | 2,515 | 1,214 | -52% | 0 | 0 | — |
case-14 | fail→pass | 11,298 | 2,222 | -80% | 1 | 1 | 0% | 1,779 | 1,156 | -35% | 0 | 0 | — |
case-15 | pass→pass | 14,736 | 2,017 | -86% | 1 | 1 | 0% | 2,385 | 1,088 | -54% | 0 | 0 | — |
case-20 | pass→pass | 10,208 | 4,084 | -60% | 1 | 1 | 0% | 1,706 | 1,483 | -13% | 0 | 0 | — |
case-16 | fail→pass | 10,479 | 5,119 | -51% | 1 | 1 | 0% | 1,695 | 1,650 | -3% | 0 | 0 | — |
case-17 | fail→pass | 16,757 | 4,645 | -72% | 1 | 1 | 0% | 3,619 | 1,696 | -53% | 0 | 0 | — |
case-18 | fail→pass | 12,598 | 4,089 | -68% | 1 | 1 | 0% | 1,991 | 1,408 | -29% | 0 | 0 | — |
case-19 | fail→pass | 14,644 | 1,847 | -87% | 1 | 1 | 0% | 2,355 | 1,063 | -55% | 0 | 0 | — |
case-21 | pass→pass | 13,994 | 7,984 | -43% | 1 | 1 | 0% | 1,972 | 2,016 | +2% | 0 | 0 | — |
case-22 | fail→pass | 11,371 | 6,180 | -46% | 1 | 1 | 0% | 1,826 | 1,864 | +2% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +59 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.