Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Design concrete removal steps for a constraint with timeline and resource needs.
.claude/skills/yogsoth-ai-removal-path/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 93% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 39% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 67% | 0% |
Design a concrete, actionable path to remove or mitigate a constraint. Produces sequenced steps with timeline, resource requirements, and success criteria for each step.
Spawns a subagent that:
Removal path design requires creative problem-solving and detailed planning that benefits from focused attention without distraction from other assessment tasks.
Output MUST include: at least 3 sequenced steps, timeline estimate, resource requirements, and success criteria for each step. Reject if steps are vague or unactionable.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | spawn-agent | Spawn a customized CC subagent with full MCP tool access. Used by SOPs that declare execution: subagent. |
<!-- 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 | 24,976 | 23,733 | -5% | 1 | 1 | 0% | 3,813 | 4,010 | +5% | 0 | 0 | — |
case-02 | fail→pass | 27,501 | 29,233 | +6% | 1 | 1 | 0% | 2,524 | 4,865 | +93% | 0 | 0 | — |
case-03 | pass→pass | 20,525 | 35,801 | +74% | 1 | 1 | 0% | 2,508 | 5,374 | +114% | 0 | 0 | — |
case-04 | fail→pass | 26,195 | 24,380 | -7% | 1 | 1 | 0% | 3,213 | 3,425 | +7% | 0 | 0 | — |
case-05 | fail→pass | 24,952 | 34,550 | +38% | 1 | 1 | 0% | 3,768 | 5,223 | +39% | 0 | 0 | — |
case-06 | fail→pass | 24,834 | 36,189 | +46% | 1 | 1 | 0% | 3,016 | 5,032 | +67% | 0 | 0 | — |
case-07 | fail→pass | 26,135 | 26,311 | +1% | 1 | 1 | 0% | 3,225 | 3,719 | +15% | 0 | 0 | — |
case-08 | fail→pass | 29,625 | 28,488 | -4% | 1 | 1 | 0% | 3,985 | 4,096 | +3% | 0 | 0 | — |
case-09 | fail→pass | 29,422 | 22,268 | -24% | 1 | 1 | 0% | 3,746 | 2,911 | -22% | 0 | 0 | — |
case-10 | fail→pass | 18,197 | 23,436 | +29% | 1 | 1 | 0% | 2,698 | 3,050 | +13% | 0 | 0 | — |
case-11 | fail→pass | 27,711 | 31,457 | +14% | 1 | 1 | 0% | 3,278 | 5,278 | +61% | 0 | 0 | — |
case-12 | fail→pass | 37,204 | 18,661 | -50% | 1 | 1 | 0% | 2,814 | 3,106 | +10% | 0 | 0 | — |
case-13 | fail→pass | 24,637 | 24,688 | +0% | 1 | 1 | 0% | 2,906 | 3,320 | +14% | 0 | 0 | — |
case-14 | fail→pass | 23,843 | 33,918 | +42% | 1 | 1 | 0% | 2,906 | 4,587 | +58% | 0 | 0 | — |
case-15 | fail→pass | 28,977 | 31,274 | +8% | 1 | 1 | 0% | 3,905 | 4,609 | +18% | 0 | 0 | — |
case-16 | fail→pass | 30,195 | 26,424 | -12% | 1 | 1 | 0% | 3,971 | 3,701 | -7% | 0 | 0 | — |
case-17 | fail→pass | 24,849 | 33,997 | +37% | 1 | 1 | 0% | 3,175 | 5,860 | +85% | 0 | 0 | — |
case-18 | fail→pass | 36,437 | 37,276 | +2% | 1 | 1 | 0% | 4,703 | 5,352 | +14% | 0 | 0 | — |
case-19 | fail→pass | 32,460 | 39,236 | +21% | 1 | 1 | 0% | 5,201 | 5,992 | +15% | 0 | 0 | — |
case-20 | pass→fail | 22,125 | 63,630 | +188% | 1 | 1 | 0% | 2,507 | 5,904 | +136% | 0 | 0 | — |
case-21 | pass→pass | 23,418 | 22,340 | -5% | 1 | 1 | 0% | 3,041 | 3,574 | +18% | 0 | 0 | — |
case-22 | pass→fail | 11,181 | 17,361 | +55% | 1 | 1 | 0% | 1,096 | 2,306 | +110% | 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 +73 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 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.