Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Strategy for iterative ontology improvement — merge duplicates, fill gaps, update confidence, prune dead branches.
.claude/skills/yogsoth-ai-ontology-refinement/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 905% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -49% | 0% |
Iteratively improve the ontology. Merge near-duplicates, fill identified gaps, update confidence scores based on new evidence, prune concepts that proved irrelevant.
Refinement is where ontologies become useful. The first pass is always rough — refinement makes it precise. Be willing to delete concepts that don't earn their place.
| Metric | S | M | L | |--------|---|---|---| | Merges performed | 2 | 5 | 12 | | Gaps filled | 3 | 8 | 15 | | Confidence updates | 5 | 15 | 30 |
| Metric | Target | Current | Status |
|---------------------|--------|---------|--------|
| Merges performed | X | 0 | ⬜ |
| Gaps filled | X | 0 | ⬜ |
| Confidence updates | X | 0 | ⬜ |<HARD-GATE> Cannot exit until 80% of budget met. Print state ledger before each iteration decision. </HARD-GATE>
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | concept-decomposition | Tactic for breaking compound concepts into atomic parts — split over-broad concepts, identify sub-components, create child pages. | | knowledge-structuring-consistency-checking | Tactic for verifying ontology consistency — detect contradictions, cycles, orphans, and type violations. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | alias-resolution | SOP for detecting and resolving concept aliases — merge duplicate pages, redirect edges. | | confidence-update | SOP for updating confidence scores on claims and evidence pages based on new information. | | gap-detection | SOP for finding structural gaps in the ontology — missing concepts, thin branches, disconnected clusters. | | merge-candidates | SOP for identifying near-duplicate concepts that should be merged. | | ontology-export | SOP for exporting ontology summary — generate a readable overview of the current ontology state. |
<!-- 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 | pass→pass | 43,423 | 42,490 | -2% | 1 | 1 | 0% | 8,285 | 7,936 | -4% | 0 | 0 | — |
case-02 | fail→pass | 41,911 | 48,879 | +17% | 1 | 1 | 0% | 6,947 | 8,806 | +27% | 0 | 0 | — |
case-03 | pass→pass | 30,378 | 38,193 | +26% | 1 | 1 | 0% | 4,696 | 6,526 | +39% | 0 | 0 | — |
case-04 | pass→fail | 22,463 | 38,352 | +71% | 1 | 1 | 0% | 3,985 | 7,604 | +91% | 0 | 0 | — |
case-05 | pass→fail | 7,820 | 34,384 | +340% | 1 | 1 | 0% | 1,576 | 6,348 | +303% | 0 | 0 | — |
case-06 | pass→fail | 32,472 | 29,242 | -10% | 1 | 1 | 0% | 2,659 | 6,541 | +146% | 0 | 0 | — |
case-07 | pass→pass | 10,080 | 12,739 | +26% | 1 | 1 | 0% | 1,546 | 2,831 | +83% | 0 | 0 | — |
case-08 | fail→pass | 12,019 | 9,897 | -18% | 1 | 1 | 0% | 1,989 | 1,381 | -31% | 0 | 0 | — |
case-09 | fail→pass | 42,813 | 50,411 | +18% | 1 | 1 | 0% | 766 | 7,700 | +905% | 0 | 0 | — |
case-10 | fail→fail | 12,030 | 5,140 | -57% | 1 | 1 | 0% | 1,886 | 1,509 | -20% | 0 | 0 | — |
case-11 | fail→fail | 7,746 | 5,607 | -28% | 1 | 1 | 0% | 1,292 | 1,595 | +23% | 0 | 0 | — |
case-12 | pass→pass | 11,174 | 32,565 | +191% | 1 | 1 | 0% | 1,755 | 5,629 | +221% | 0 | 0 | — |
case-13 | fail→pass | 11,086 | 7,652 | -31% | 1 | 1 | 0% | 1,993 | 1,884 | -5% | 0 | 0 | — |
case-14 | pass→pass | 23,195 | 20,074 | -13% | 1 | 1 | 0% | 1,992 | 4,456 | +124% | 0 | 0 | — |
case-15 | pass→pass | 12,887 | 3,057 | -76% | 1 | 1 | 0% | 2,051 | 1,059 | -48% | 0 | 0 | — |
case-16 | fail→pass | 12,425 | 2,789 | -78% | 1 | 1 | 0% | 1,979 | 1,015 | -49% | 0 | 0 | — |
case-17 | pass→pass | 10,658 | 2,618 | -75% | 1 | 1 | 0% | 1,704 | 991 | -42% | 0 | 0 | — |
case-18 | fail→pass | 11,786 | 2,241 | -81% | 1 | 1 | 0% | 1,854 | 922 | -50% | 0 | 0 | — |
case-19 | fail→pass | 10,768 | 10,096 | -6% | 1 | 1 | 0% | 1,677 | 2,355 | +40% | 0 | 0 | — |
case-20 | pass→pass | 12,574 | 13,102 | +4% | 1 | 1 | 0% | 1,886 | 2,508 | +33% | 0 | 0 | — |
case-21 | fail→pass | 7,166 | 10,004 | +40% | 1 | 1 | 0% | 1,149 | 2,264 | +97% | 0 | 0 | — |
case-22 | pass→pass | 39,004 | 1,902 | -95% | 1 | 1 | 0% | 3,051 | 933 | -69% | 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 +23 percentage points is the difference between those two pass rates over the 21 comparable cases. 3 cases got worse with the skill loaded, and they are 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.