Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Determine when additional searching yields diminishing returns. Analyzes the latest expansion batch against existing corpus to judge continue/near-saturation/saturated. Used by snowball and systematic-survey.
.claude/skills/yogsoth-ai-knowledge-acquisition-saturation-detection/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 47% | 0% |
| case-20 | ✓→✓ | ＝ Same ✓ | 36% | 0% |
| case-21 | ✓→✓ | ＝ Same ✓ | 284% | 0% |
| case-22 | ✓→✓ | ＝ Same ✓ | 10% | 0% |
| case-04 | ✗→✗ | ＝ Same ✗ | 181% | 0% |
Determine when to stop expanding — diminishing returns analysis.
Subagent — spawned via subagent-spawning/spawn-agent skill.
Saturation judgment requires comparing the entire existing corpus against the latest batch of new papers. The comparison is context-intensive and benefits from dedicated processing.
<!-- 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-04 | fail→fail | 11,770 | 35,095 | +198% | 1 | 1 | 0% | 988 | 2,778 | +181% | 0 | 0 | — |
case-05 | fail→fail | 15,580 | 33,121 | +113% | 1 | 1 | 0% | 1,667 | 4,703 | +182% | 0 | 0 | — |
case-01 | fail→fail | 20,666 | 12,961 | -37% | 1 | 1 | 0% | 2,423 | 1,217 | -50% | 0 | 0 | — |
case-02 | fail→fail | 12,299 | 33,659 | +174% | 1 | 1 | 0% | 1,168 | 4,686 | +301% | 0 | 0 | — |
case-03 | fail→fail | 17,751 | 37,339 | +110% | 1 | 1 | 0% | 1,822 | 3,852 | +111% | 0 | 0 | — |
case-06 | fail→fail | 18,043 | 19,027 | +5% | 1 | 1 | 0% | 2,054 | 2,457 | +20% | 0 | 0 | — |
case-07 | fail→fail | 23,834 | 45,760 | +92% | 1 | 1 | 0% | 2,852 | 3,952 | +39% | 0 | 0 | — |
case-08 | fail→fail | 22,148 | 34,522 | +56% | 1 | 1 | 0% | 2,697 | 4,849 | +80% | 0 | 0 | — |
case-09 | fail→fail | 17,131 | 17,573 | +3% | 1 | 1 | 0% | 1,814 | 2,160 | +19% | 0 | 0 | — |
case-10 | fail→fail | 17,792 | 29,808 | +68% | 1 | 1 | 0% | 1,987 | 4,086 | +106% | 0 | 0 | — |
case-11 | fail→fail | 16,576 | 26,969 | +63% | 1 | 1 | 0% | 1,672 | 3,815 | +128% | 0 | 0 | — |
case-12 | fail→fail | 21,646 | 34,818 | +61% | 1 | 1 | 0% | 2,479 | 5,085 | +105% | 0 | 0 | — |
case-13 | fail→fail | 23,240 | 30,137 | +30% | 1 | 1 | 0% | 2,730 | 4,308 | +58% | 0 | 0 | — |
case-14 | fail→pass | 18,087 | 31,562 | +75% | 1 | 1 | 0% | 1,839 | 2,705 | +47% | 0 | 0 | — |
case-15 | fail→fail | 23,591 | 37,025 | +57% | 1 | 1 | 0% | 2,973 | 5,592 | +88% | 0 | 0 | — |
case-16 | fail→fail | 21,477 | 38,284 | +78% | 1 | 1 | 0% | 2,548 | 3,757 | +47% | 0 | 0 | — |
case-17 | fail→fail | 20,536 | 27,109 | +32% | 1 | 1 | 0% | 2,288 | 3,912 | +71% | 0 | 0 | — |
case-18 | fail→fail | 16,738 | 29,134 | +74% | 1 | 1 | 0% | 1,792 | 4,110 | +129% | 0 | 0 | — |
case-19 | fail→fail | 26,083 | 35,686 | +37% | 1 | 1 | 0% | 3,186 | 5,269 | +65% | 0 | 0 | — |
case-20 | pass→pass | 14,773 | 16,072 | +9% | 1 | 1 | 0% | 1,847 | 2,510 | +36% | 0 | 0 | — |
case-21 | pass→pass | 6,916 | 11,772 | +70% | 1 | 1 | 0% | 301 | 1,156 | +284% | 0 | 0 | — |
case-22 | pass→pass | 20,080 | 19,579 | -2% | 1 | 1 | 0% | 2,586 | 2,855 | +10% | 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 +5 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.