council-of-minds
Convene 6 intellectual personas across multiple LLM providers for hard decisions. Each persona runs on a different provider for genuinely different reasoning — not costume changes on one model. Inspired by council-of-high-intelligence (CC0).
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 13
- Forks
- 2
01 前言 Front Matter
- name
- council-of-minds
- modes
- { "duo": "2 opposing personas, 1 round (~45s)", "full": "6 personas, 2 rounds (~4 min)", "quick": "3 personas, 1 round (~90s)" }
- origin
- yana-ai — inspired by 0xNyk/council-of-high-intelligence (CC0)
- license
- MIT
- version
- 1.0.0
- triggers
- [ "/council-of-minds", "council of minds", "convene the council", "multi-mind deliberation" ]
- description
- Convene 6 intellectual personas across multiple LLM providers for hard decisions. Each persona runs on a different provider for genuinely different reasoning — not costume changes on one model. Inspired by council-of-high-intelligence (CC0).
03 SKILL.md 正文
/council-of-minds
Six intellectual personas deliberate your hardest decisions across multiple AI providers. One command. Structured disagreement. No fake consensus.
Modes
/council-of-minds "question" → full (6 personas, 2 rounds)
/council-of-minds --quick "question" → quick (3 personas, 1 round)
/council-of-minds --duo "question" → duo (2 opposing voices, 1 round)
When to Use
- Decision has ≥2 credible paths and no obvious winner
- You want structured disagreement, not a polished "yes"
- You suspect your own anchoring bias
- High stakes: architecture, strategy, go/no-go, hiring, product direction
When NOT to Use
- Single correct answer (use tools or research instead)
- Pure implementation task (use
/plan) - Code review (use
/code-review) - Factual lookup (just ask directly)
The Six Personas
Each persona is routed to a different provider to ensure genuinely different reasoning.
| Persona | Intellectual Tradition | Provider | Lens |
|---|---|---|---|
| The Architect | Systems thinking, long-term | Claude | Correctness, maintainability, 5-year consequence |
| Socrates | Socratic questioning | Gemini | Destroys premises, finds the hidden assumption |
| Feynman | First-principles empiricism | DeepSeek | "What does evidence actually say? Rebuild from scratch." |
| The Pragmatist | Execution realism | Groq | Shipping speed, operational reality, resource constraints |
| Kahneman | Behavioral economics | OpenAI | Cognitive bias traps, System 1 vs System 2, loss aversion |
| Torvalds | Engineering minimalism | Ollama (local) | "Will this actually work? Is it unnecessarily complex?" |
If a provider is unavailable, fall back to Claude for that persona but explicitly note the fallback in output.
Protocol
Step 0 — Problem Restate Gate
Before any analysis, every persona must reframe the question in one sentence.
Rule: if ≥3 personas restate the question differently from how it was asked, the original question was the problem. Resolve the question ambiguity before proceeding.
Step 1 — Extract the real decision
Reduce to:
- What exactly are we deciding?
- What constraints are non-negotiable?
- What does success look like in 6 months?
- What is the cost of reversing this decision?
Step 2 — Form Architect position first (anti-anchoring)
Before reading other voices, write:
- Initial position (1 sentence)
- Three strongest reasons
- Main risk in preferred path
This prevents synthesis from simply mirroring external voices.
Step 3 — Launch personas in parallel
Each persona gets only:
- The restated decision question
- Minimal necessary context (no full conversation history)
- Their role prompt
Persona prompt template:
You are [PERSONA NAME] on a six-voice decision council.
Your intellectual tradition: [tradition]
Your primary lens: [lens]
Question: [restated decision]
Context (only what's needed):
[compact context — code snippet, constraint, metric — max 200 words]
Respond with exactly this structure:
1. Restate — rephrase the question in your own words (1 sentence)
2. Position — your recommendation (1-2 sentences)
3. Reasoning — 3 bullets, each ≤ 25 words
4. Strongest objection to your own position — 1 sentence
5. What the other voices will miss — 1 sentence
Under 250 words total. No hedging. No "it depends" without specifying what it depends on.
Step 4 — Dissent Quota Check
After collecting all positions:
Rule: If ≥4 of 6 personas agree on the same recommendation, force 2 personas (chosen by least-confident alignment) to steelman the opposing view before synthesis.
This prevents groupthink from producing false consensus.
Step 5 — Round 2 (full mode only)
Show each persona the other five positions (anonymized — no names, just positions). Each updates or defends their view in ≤ 100 words.
Step 6 — Synthesize
Rules for synthesis:
- Do not dismiss any external view without explaining why
- If an external voice changed the Architect's recommendation, say so explicitly
- Lead with Unresolved Questions before the verdict
- The strongest dissent must appear even if rejected
Output Format
## Council of Minds: [decision title]
### Problem Restate Gate
- Original question: [as asked]
- Restated by council: [if different, flag it]
---
### Positions
**The Architect** (Claude) — [1-sentence position]
> [key reasoning, 1 line]
**Socrates** (Gemini) — [1-sentence position]
> [key reasoning, 1 line]
**Feynman** (DeepSeek) — [1-sentence position]
> [key reasoning, 1 line]
**The Pragmatist** (Groq) — [1-sentence position]
> [key reasoning, 1 line]
**Kahneman** (OpenAI) — [1-sentence position]
> [key reasoning, 1 line]
**Torvalds** (Ollama) — [1-sentence position]
> [key reasoning, 1 line]
---
### Synthesis
**Where they align:** [what ≥4 agree on]
**Strongest dissent:** [the most important disagreement — do not soften it]
**Dissent quota triggered:** yes / no — [if yes, who steelmanned what]
**Unresolved questions:**
- [question the council couldn't answer]
- [what you'd need to know to decide with confidence]
**Recommended next steps:** [concrete, not "do more research"]
### Verdict
**Recommendation:** [the synthesized path — commit to one]
**Confidence:** [low / medium / high + why]
**Reversibility:** [how hard to undo if wrong]
Quick Mode (3 personas)
Use: Architect + Socrates + Torvalds
Best for: tactical decisions, scope calls, technology choices
One round only. No Round 2. No dissent quota.
Output: same format but 3 positions only.
Duo Mode (2 personas)
Pick the two most opposing intellectual traditions for the question type:
| Decision type | Best duo |
|---|---|
| Technical architecture | Architect vs Torvalds |
| Strategic direction | Feynman vs Kahneman |
| Ship vs hold | Pragmatist vs Architect |
| Question a premise | Socrates vs anyone |
One round. No restate gate (too short). Verdict required.
Anti-Patterns
❌ Feeding the full conversation history to all personas (kills independence)
❌ Hiding disagreement in the verdict to sound decisive
❌ Using council for tasks with an objectively correct answer
❌ Accepting ≥4-way agreement without triggering dissent quota
❌ "It depends" without specifying the dependency
❌ Council when the real blocker is missing information, not conflicting values
Related Skills
council— 4-voice version, lightermulti-agent-debate— Du et al. 2023 algorithm, multi-round consensussanta-method— adversarial verification of a specific claimfirst-principles-thinker— single voice, Feynman-style deconstruction
04 同仓库更多 Skills
openai--sentry--sentry
Use when the user asks to inspect Sentry issues or events, summarize recent production errors, or pull basic Sentry health data via the Sentry API; perform read-only queries with the bundled script and require `SENTRY_AUTH_TOKEN`.
ito-basket-compare
Compare Itô prediction-market baskets against a user's knowledge base, portfolio notes, financial context, watchlist, or research thesis. Use for read-only basket comparison and gap analysis without investment advice or live trading.
implementing-network-traffic-baselining
Build network traffic baselines from NetFlow/IPFIX data using Python pandas for statistical analysis, z-score anomaly detection, and hourly/daily traffic pattern profiling
detecting-evasion-techniques-in-endpoint-logs
Detects defense evasion techniques used by adversaries in endpoint logs including log tampering, timestomping, process injection, and security tool disabling. Use when investigating suspicious endpoint behavior, building detection rules for evasion tactics, or conducting threat hunting for stealthy adversary activity. Activates for requests involving evasion detection, defense evasion analysis, log tampering detection, or MITRE ATT&CK TA0005.
implementing-microsegmentation-with-guardicore
Implementing microsegmentation using Akamai Guardicore Segmentation to map application dependencies, create granular network policies, visualize east-west traffic flows, and enforce least-privilege communication between workloads across data centers and cloud.
ramdisk-tmpfs-patterns
暂无简介