Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Tactic for breaking compound concepts into atomic parts — split over-broad concepts, identify sub-components, create child pages.
.claude/skills/yogsoth-ai-concept-decomposition/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -24% | 0% |
Break compound or over-broad concepts into their atomic constituents. A concept that means too many things should become multiple focused concepts connected by edges.
<HARD-GATE> ≥2 child concepts created per decomposition invocation. If a concept cannot be meaningfully decomposed, report it as atomic and exit. </HARD-GATE>
<!-- BEGIN available-tables (generated) -->
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. | | concept-page-creation | SOP for creating a new concept page with proper frontmatter, content, and initial edges. | | edge-batch-creation | SOP for creating multiple edges in a batch — efficient bulk relationship creation. |
<!-- 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-06 | fail→pass | 12,758 | 4,437 | -65% | 1 | 1 | 0% | 1,985 | 1,048 | -47% | 0 | 0 | — |
case-01 | fail→pass | 18,359 | 14,795 | -19% | 1 | 1 | 0% | 3,044 | 2,944 | -3% | 0 | 0 | — |
case-02 | fail→pass | 13,777 | 8,514 | -38% | 1 | 1 | 0% | 2,391 | 1,869 | -22% | 0 | 0 | — |
case-03 | fail→pass | 13,452 | 13,568 | +1% | 1 | 1 | 0% | 2,218 | 2,833 | +28% | 0 | 0 | — |
case-04 | pass→pass | 10,527 | 8,236 | -22% | 1 | 1 | 0% | 1,698 | 1,703 | +0% | 0 | 0 | — |
case-05 | pass→pass | 11,796 | 4,666 | -60% | 1 | 1 | 0% | 1,914 | 1,102 | -42% | 0 | 0 | — |
case-07 | fail→pass | 12,382 | 6,233 | -50% | 1 | 1 | 0% | 1,819 | 1,375 | -24% | 0 | 0 | — |
case-08 | pass→pass | 5,905 | 3,941 | -33% | 1 | 1 | 0% | 897 | 973 | +8% | 0 | 0 | — |
case-09 | fail→pass | 13,876 | 2,337 | -83% | 1 | 1 | 0% | 2,137 | 685 | -68% | 0 | 0 | — |
case-10 | pass→pass | 8,987 | 2,495 | -72% | 1 | 1 | 0% | 1,347 | 739 | -45% | 0 | 0 | — |
case-11 | fail→pass | 19,408 | 12,171 | -37% | 1 | 1 | 0% | 2,972 | 2,436 | -18% | 0 | 0 | — |
case-12 | fail→pass | 16,858 | 13,134 | -22% | 1 | 1 | 0% | 2,740 | 2,648 | -3% | 0 | 0 | — |
case-13 | pass→pass | 11,646 | 13,080 | +12% | 1 | 1 | 0% | 1,873 | 2,410 | +29% | 0 | 0 | — |
case-14 | fail→fail | 14,737 | 10,097 | -31% | 1 | 1 | 0% | 2,184 | 1,746 | -20% | 0 | 0 | — |
case-15 | fail→pass | 10,343 | 2,397 | -77% | 1 | 1 | 0% | 1,799 | 725 | -60% | 0 | 0 | — |
case-16 | fail→pass | 18,590 | 11,822 | -36% | 1 | 1 | 0% | 2,943 | 2,403 | -18% | 0 | 0 | — |
case-17 | pass→fail | 10,509 | 3,023 | -71% | 1 | 1 | 0% | 1,605 | 734 | -54% | 0 | 0 | — |
case-18 | fail→pass | 11,692 | 2,522 | -78% | 1 | 1 | 0% | 1,644 | 775 | -53% | 0 | 0 | — |
case-19 | pass→pass | 13,528 | 9,384 | -31% | 1 | 1 | 0% | 2,489 | 1,997 | -20% | 0 | 0 | — |
case-20 | pass→pass | 20,359 | 10,143 | -50% | 1 | 1 | 0% | 3,412 | 2,091 | -39% | 0 | 0 | — |
case-21 | pass→pass | 14,946 | 6,117 | -59% | 1 | 1 | 0% | 2,725 | 1,480 | -46% | 0 | 0 | — |
case-22 | pass→pass | 5,612 | 2,151 | -62% | 1 | 1 | 0% | 724 | 631 | -13% | 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. The headline lift of +45 percentage points is the difference between those two pass rates over the 22 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.