Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →4-voice parallel deliberation (Architect · Skeptic · Pragmatist · Critic) for architecture, tech selection, or design decisions with no clear answer. Anti-anchoring: each voice gets independent context. Records decision in harness-mem.
.claude/skills/hashgraph-online-council/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 118% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 83% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 54% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 477% | 0% |
NO MAJOR DECISION WITHOUT AT LEAST TWO OPPOSING VIEWPOINTS.
State the decision in one clear question. Example: "Should we use WebSockets or SSE for real-time updates?"
Launch 4 independent subagents using the Agent tool, each with ONLY the question and relevant context (NOT the full conversation history — this prevents anchoring bias):
| Voice | Role | Focus | |-------|------|-------| | Architect | Long-term correctness | Maintainability, extensibility, architectural alignment | | Skeptic | Challenge assumptions | Simpler alternatives, hidden costs, "what if we don't?" | | Pragmatist | Ship it now | Timeline, user impact, operational complexity, team skills | | Critic | Find the cracks | Edge cases, failure modes, migration risks, rollback difficulty |
Anti-Anchoring Rule: Each voice receives only:
They do NOT receive: the full conversation, other voices' opinions, or the user's leaning.
After all 4 voices report back:
Save the decision to memory:
bashepic mem add \ --title "Decision: {question}" \ --type decision \ --importance 0.9 \ --body "Context: ...\nOptions: ...\nChosen: ...\nRationale: ...\nTrade-off accepted: ..."
| Excuse | Rebuttal | What to do instead | |--------|----------|-------------------| | "I already know the answer" | Your gut feeling is not analysis. Council surfaces blind spots. | Run Council anyway — you'll learn something. | | "This is too slow for a simple decision" | 5 minutes of council prevents weeks of regret. | Set a 5-minute timer and run the full process. | | "The team already agreed" | Groupthink is not consensus. Diverse perspectives prevent disasters. | Summon all 4 voices — especially the Skeptic and Critic. | | "I don't need 4 voices for this" | Even simple decisions benefit from the Skeptic and Critic. | Run all 4 voices. Skipping voices skips insight. |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 32,252 | 46,437 | +44% | 1 | 1 | 0% | 3,517 | 7,682 | +118% | 0 | 0 | — |
case-11 | fail→pass | 12,475 | 35,902 | +188% | 1 | 1 | 0% | 2,459 | 4,497 | +83% | 0 | 0 | — |
case-12 | fail→pass | 17,083 | 32,510 | +90% | 1 | 1 | 0% | 2,809 | 5,843 | +108% | 0 | 0 | — |
case-22 | pass→pass | 37,834 | 39,322 | +4% | 1 | 1 | 0% | 5,035 | 6,882 | +37% | 0 | 0 | — |
case-02 | fail→fail | 29,723 | 18,707 | -37% | 1 | 1 | 0% | 4,116 | 1,184 | -71% | 0 | 0 | — |
case-03 | fail→fail | 38,638 | 26,320 | -32% | 1 | 1 | 0% | 5,832 | 2,477 | -58% | 0 | 0 | — |
case-04 | pass→pass | 7,767 | 18,497 | +138% | 1 | 1 | 0% | 1,152 | 2,320 | +101% | 0 | 0 | — |
case-05 | pass→fail | 5,873 | 18,155 | +209% | 1 | 1 | 0% | 1,031 | 1,083 | +5% | 0 | 0 | — |
case-06 | pass→fail | 3,705 | 11,661 | +215% | 1 | 1 | 0% | 705 | 1,171 | +66% | 0 | 0 | — |
case-07 | fail→pass | 25,168 | 47,102 | +87% | 1 | 1 | 0% | 3,317 | 5,097 | +54% | 0 | 0 | — |
case-08 | fail→fail | 8,220 | 12,725 | +55% | 1 | 1 | 0% | 302 | 1,059 | +251% | 0 | 0 | — |
case-09 | fail→fail | 25,748 | 6,370 | -75% | 1 | 1 | 0% | 2,991 | 1,224 | -59% | 0 | 0 | — |
case-10 | fail→pass | 13,051 | 38,198 | +193% | 1 | 1 | 0% | 1,119 | 6,456 | +477% | 0 | 0 | — |
case-13 | fail→pass | 28,434 | 39,221 | +38% | 1 | 1 | 0% | 3,974 | 5,332 | +34% | 0 | 0 | — |
case-14 | fail→fail | 11,511 | 9,833 | -15% | 1 | 1 | 0% | 1,585 | 1,425 | -10% | 0 | 0 | — |
case-15 | fail→fail | 25,353 | 12,296 | -52% | 1 | 1 | 0% | 2,884 | 1,284 | -55% | 0 | 0 | — |
case-16 | pass→pass | 23,665 | 33,728 | +43% | 1 | 1 | 0% | 2,601 | 5,572 | +114% | 0 | 0 | — |
case-17 | pass→fail | 23,784 | 23,156 | -3% | 1 | 1 | 0% | 2,876 | 1,261 | -56% | 0 | 0 | — |
case-18 | fail→fail | 18,850 | 23,828 | +26% | 1 | 1 | 0% | 1,970 | 1,364 | -31% | 0 | 0 | — |
case-19 | pass→fail | 27,373 | 19,092 | -30% | 1 | 1 | 0% | 3,162 | 1,412 | -55% | 0 | 0 | — |
case-20 | fail→pass | 11,360 | 32,390 | +185% | 1 | 1 | 0% | 1,843 | 4,064 | +121% | 0 | 0 | — |
case-21 | fail→pass | 16,763 | 19,450 | +16% | 1 | 1 | 0% | 1,851 | 2,202 | +19% | 0 | 0 | — |
case-23 | pass→fail | 16,269 | 36,316 | +123% | 1 | 1 | 0% | 2,330 | 1,116 | -52% | 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. 23 cases were attempted, and 11 counted toward the lift figure. The other 12 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 +13 percentage points is the difference between those two pass rates over the 11 comparable cases. 6 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.