Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Strategy: 10th Man Rule and Liberating Structures — institutionalized dissent to prevent premature consensus and expose suppressed objections.
.claude/skills/yogsoth-ai-groupthink-mitigation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 73% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 119% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 39% | 0% |
Prevent premature consensus by mandating structured dissent. Based on Israeli intelligence 10th Man doctrine and Lipmanowicz Liberating Structures.
| Parameter | S | M | L | |---|---|---|---| | Attack vectors | 5 | 12 | 20 | | Probing rounds | 3 | 6 | 10 | | Personas | 2 | 4 | 6 | | Assumption checks | 5 | 10 | 20 |
persona-construction → [build 10th Man dissenter]
→ devils-advocacy (construct dissenting case)
→ [for each dissenting claim]:
probe-execution (test claim)
→ key-assumptions-check (verify consensus assumptions)
→ finding-aggregation → attack-resilience-scoring<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | adversarial-roleplay | Tactic: Construct detailed hostile persona, attack artifact from that persona's perspective, record successful attack paths for aggregation. | | assumption-cascade | Tactic: Surface assumptions, sort by dependency, attack root assumptions first, then trace cascade failures through the dependency graph. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | devils-advocacy | Construct the strongest possible counter-argument against a position, steelmanning the opposition before attacking. | | finding-aggregation | Aggregate, deduplicate, and classify findings from multiple probes into a coherent vulnerability report. | | key-assumptions-check | Military ACT: systematically enumerate all assumptions, classify by type, and evaluate evidence strength supporting each. | | persona-construction | Build a detailed adversarial persona with background, motivation, expertise, blind spots, and preferred attack patterns. | | probe-execution | Execute a single attack probe against an artifact, record the result with evidence and severity classification. |
<!-- 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 | 27,217 | 38,275 | +41% | 1 | 1 | 0% | 3,574 | 4,529 | +27% | 0 | 0 | — |
case-02 | fail→pass | 29,533 | 30,657 | +4% | 1 | 1 | 0% | 3,951 | 4,862 | +23% | 0 | 0 | — |
case-03 | fail→pass | 21,361 | 32,628 | +53% | 1 | 1 | 0% | 2,988 | 5,169 | +73% | 0 | 0 | — |
case-04 | pass→fail | 10,799 | 23,135 | +114% | 1 | 1 | 0% | 1,618 | 3,989 | +147% | 0 | 0 | — |
case-05 | fail→fail | 2,811 | 7,076 | +152% | 1 | 1 | 0% | 377 | 1,535 | +307% | 0 | 0 | — |
case-06 | fail→fail | 4,600 | 6,562 | +43% | 1 | 1 | 0% | 631 | 1,524 | +142% | 0 | 0 | — |
case-07 | fail→pass | 11,846 | 18,361 | +55% | 1 | 1 | 0% | 1,461 | 3,205 | +119% | 0 | 0 | — |
case-08 | fail→pass | 26,187 | 27,833 | +6% | 1 | 1 | 0% | 3,247 | 4,526 | +39% | 0 | 0 | — |
case-09 | fail→pass | 22,018 | 39,250 | +78% | 1 | 1 | 0% | 2,924 | 6,109 | +109% | 0 | 0 | — |
case-18 | pass→pass | 19,945 | 30,865 | +55% | 1 | 1 | 0% | 2,858 | 5,177 | +81% | 0 | 0 | — |
case-10 | pass→pass | 16,294 | 31,932 | +96% | 1 | 1 | 0% | 2,215 | 4,053 | +83% | 0 | 0 | — |
case-11 | pass→pass | 20,098 | 34,718 | +73% | 1 | 1 | 0% | 2,728 | 5,593 | +105% | 0 | 0 | — |
case-12 | pass→pass | 18,281 | 43,423 | +138% | 1 | 1 | 0% | 2,444 | 6,822 | +179% | 0 | 0 | — |
case-13 | fail→pass | 16,228 | 14,122 | -13% | 1 | 1 | 0% | 2,411 | 2,709 | +12% | 0 | 0 | — |
case-14 | fail→pass | 12,564 | 5,917 | -53% | 1 | 1 | 0% | 1,718 | 1,654 | -4% | 0 | 0 | — |
case-15 | pass→pass | 19,087 | 27,063 | +42% | 1 | 1 | 0% | 2,593 | 4,384 | +69% | 0 | 0 | — |
case-16 | pass→pass | 18,270 | 33,492 | +83% | 1 | 1 | 0% | 2,319 | 5,293 | +128% | 0 | 0 | — |
case-17 | pass→pass | 16,142 | 21,645 | +34% | 1 | 1 | 0% | 2,264 | 3,518 | +55% | 0 | 0 | — |
case-19 | pass→pass | 18,335 | 27,058 | +48% | 1 | 1 | 0% | 2,449 | 2,491 | +2% | 0 | 0 | — |
case-20 | pass→pass | 12,198 | 9,622 | -21% | 1 | 1 | 0% | 1,727 | 1,980 | +15% | 0 | 0 | — |
case-21 | pass→pass | 12,149 | 8,859 | -27% | 1 | 1 | 0% | 1,603 | 1,998 | +25% | 0 | 0 | — |
case-22 | pass→pass | 15,241 | 11,370 | -25% | 1 | 1 | 0% | 2,118 | 2,214 | +5% | 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 +32 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.