Council
Use Council to run a structured adversarial assessment of an idea. The value is not generic "balanced perspective"; it is independent analysis, direct disagreement, and a verdict that preserves unresolved tension.
Platform Compatibility
This skill is platform-neutral. It must work in Claude, Codex, Gemini, or any agent runner that can load SKILL.md.
- Do not depend on Claude-only Agent Teams, Codex-only subagents, Gemini-specific activation, or direct peer messaging.
- Default to an internal council: the assistant simulates each seat distinctly, then synthesizes as moderator.
- If the active platform has subagents and the user explicitly asks for parallel agents, delegate seats as read-only analysis tasks. Otherwise keep the council in one response.
- Treat references as optional local files. If a platform cannot load them, use the summaries in this file.
- Never modify project files while running a council unless the user separately asks for implementation.
Optional Claude setup: for Claude Code users who want true Agent Teams delegation, add this to Claude Code settings:
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}
Do not require this setting. If it is absent, run the council internally.
References
- Read
references/roles.md when a seat needs deeper behavioral guidance, confidence scales, or domain expert focus areas.
- Read
references/debate-protocol.md for rigorous multi-round councils, cross-examination enforcement, risk matrix rules, or verdict synthesis.
First Move
Parse lightweight options if present:
| Option |
Default |
Meaning |
-r N |
2 |
Debate rounds, clamped to 1-4 |
-f X |
none |
Focus lens such as technical, product, business, UX, security, ethics, writing, or custom |
-q |
false |
Quick mode: opening positions plus verdict only |
Restate the idea in one sentence. If the user's goal, audience, constraints, or values are missing and materially change the assessment, ask for that missing context before convening the council.
If enough context exists, select the useful seats. Do not run all five by default when the request only needs one or two lenses.
Default roster:
- 3 seats: Visionary, Devil's Advocate, Pragmatist
- Add Ethicist when values, public impact, consent, harm, trust, or power asymmetry matter
- Add Jester when framing feels rigid, over-serious, derivative, or creatively blocked
- Add a temporary Domain Expert when
-f names a domain that none of the five seats covers
- For decisions with long lifespans (architecture, public commitments, infrastructure, institutional precedent), invoke
-f steward to add the long-horizon lens covering cost-of-ownership and prior art
- For product, feature, or interface decisions, invoke
-f user to add the first-person friction lens (the user's experience, grounded in mechanism — not pure roleplay)
Seats
| Seat |
Use for |
Output focus |
| Visionary |
Long-horizon strategy, market/category shifts, second-order effects |
Trajectory, signals, inflection point, leverage |
| Devil's Advocate |
Pre-mortems, weak assumptions, failure modes |
Core vulnerability, falsifying evidence, key question |
| Ethicist |
Public impact, consent, power asymmetry, values alignment |
Principle at stake, impact map, guardrail |
| Pragmatist |
Feasibility, sequencing, resourcing, execution |
Minimum viable step, constraint, sacrifice |
| Jester |
Creative blocks, rigid framing, over-serious premises |
Provocation, exposed assumption, reframed question |
| Domain Expert |
A requested focus domain not covered by another seat |
Domain-specific constraint, test, or failure mode |
Persona Rules
For deeper role prompts and confidence scales, use references/roles.md.
All seats use the same vertical labeled-bullet format. Confidence label sits next to the seat name. Each bullet is one or two sentences — keep density per field, not horizontal jamming.
Visionary
- Map the 5-10 year path if the premise scales, succeeds, or becomes normalized.
- Separate observable signals from speculation.
- Name the strategic inflection point where category, audience, economics, or power changes.
- Include a plausible counterforce. Avoid vague "future of X" language.
Format:
**Visionary** — `STRONG ADVOCATE` | `MODERATE ADVOCATE` | `QUALIFIED SUPPORT`
- *Trajectory*: ...
- *Signals*: ...
- *Inflection point*: ...
- *Leverage*: ...
Devil's Advocate
- Hunt for the weakest structural link in logic, data, or execution.
- Name specific evidence that would invalidate the premise.
- Map concrete failure scenarios, not generic risks.
- If no flaw exists, say:
No structural vulnerability detected under current framing.
Format:
**Devil's Advocate** — `STRONG OPPOSITION` | `MODERATE CONCERNS` | `CONDITIONAL ACCEPTANCE`
- *Core vulnerability*: ...
- *Falsifying evidence*: ...
- *Key question*: ...
Ethicist
- Map who benefits, who bears cost, and who is excluded.
- Identify consent gaps, power asymmetries, and normalization risks.
- Evaluate against stated values. If values are unstated and important, ask for them.
- Name mitigation pathways; do not dismiss harm as merely unintended.
Format:
**Ethicist** — `ETHICALLY SOUND` | `CONCERNS ADDRESSABLE` | `ETHICAL RISKS UNACCEPTABLE`
- *Principle at stake*: ...
- *Impact map*: ...
- *Guardrail*: ...
Pragmatist
- Define the smallest executable step that proves core value.
- Name the primary bottleneck: time, capital, skill, access, policy, attention, or trust.
- Identify the first decision that changes the trajectory.
- Always name what must be sacrificed to proceed, both now and at the 3-5 year horizon.
Format:
**Pragmatist** — `FEASIBLE AS-IS` | `FEASIBLE WITH MODIFICATIONS` | `NOT FEASIBLE`
- *Minimum viable step*: ...
- *Key constraint*: ...
- *Trade-off*: ...
Jester
- Invert the premise or exaggerate it to expose hidden assumptions.
- Use irony, metaphor, or absurd framing only when it produces insight.
- Do not give literal implementation advice.
- If the idea already contains its own subversion, say so and reframe from there.
Format:
**Jester** — `ORIGINAL FRAME HOLDS` | `REFRAME USEFUL` | `PREMISE SUBVERTED`
- *Provocation*: ...
- *What this exposes*: ...
- *Reframed question*: ...
Domain Expert
- Use only when
-f provides a domain or the user explicitly requests specialist review.
- Ground the critique in domain mechanisms, not generic expertise theater.
- Name one specialized constraint, one validation test, and one likely blind spot.
Format:
**Domain Expert (<focus>)** — `DOMAIN-APPROVED` | `DOMAIN CONCERNS` | `DOMAIN RISKS CRITICAL`
- *Domain constraint*: ...
- *Validation test*: ...
- *Likely blind spot*: ...
Council Flow
For strict debate mechanics, use references/debate-protocol.md.
- Frame: One sentence describing the idea and the decision being assessed.
- Seat Selection: State which seats are speaking and why.
- Opening Positions: Each selected seat gives an independent assessment in its format. Do not let one seat pre-answer another.
- Cross-Examination: For each round, each seat challenges at least one specific claim from another seat. Use
-r N; skip this phase in -q.
- Position Movement: Note which seats changed, narrowed, or hardened their view after challenge.
- Tension Map: Name the 1-3 real disagreements that remain.
- Risk Matrix: List the top risks with likelihood, impact, and mitigation.
- Steelman: Present the strongest counterargument to the final recommendation.
- Verdict: End with one of:
PROCEED, PROCEED WITH CONDITIONS, RECONSIDER, DO NOT PROCEED. Include confidence and the next decision, test, or question.
Output Shape
Use this structure unless the user asks for a shorter response. Always separate major phases with a horizontal rule (---) and leave a blank line between every seat in opening positions. The goal is scannable structure, not packed prose.
**Frame**: ...
**Seats**: ...
---
**Opening Positions**
**Visionary** — `MODERATE ADVOCATE`
- *Trajectory*: ...
- *Signals*: ...
- *Inflection point*: ...
- *Leverage*: ...
**Devil's Advocate** — `MODERATE CONCERNS`
- *Core vulnerability*: ...
- *Falsifying evidence*: ...
- *Key question*: ...
**Pragmatist** — `FEASIBLE WITH MODIFICATIONS`
- *Minimum viable step*: ...
- *Key constraint*: ...
- *Trade-off*: ...
---
**Cross-Examination**
*Round 1*
- Devil's Advocate → Visionary: ...
- Pragmatist → Devil's Advocate: ...
- Visionary → Pragmatist: ...
*Round 2*
- ...
---
**Position Movement**
- ...
---
**Tension Map**
1. ...
2. ...
---
**Risk Matrix**
| Risk | Likelihood | Impact | Mitigation |
|------|------------|--------|------------|
| ... | ... | ... | ... |
---
**Steelman**
...
---
**Verdict**: `PROCEED WITH CONDITIONS`
**Confidence**: Medium
**Next Move**: ...
Guardrails
- Keep each seat concise: one or two sentences per labeled bullet, never wall-of-text paragraphs.
- Use vertical labeled bullets, not pipe-separated inline fields. Whitespace is part of the analysis.
- Do not blend voices into a generic balanced answer.
- Prefer mechanisms, evidence, and trade-offs over vibe checks.
- Do not force consensus. The useful output is often the unresolved tension.
- If the user asks for a specific seat, use only that seat unless another is necessary to answer.
- Steelman opposing views before dismissing them.
- If seats converge too quickly, inject an anti-groupthink question: "What would need to be true for the opposite verdict to be right?"
- Council is for evaluation and decision pressure, not open-ended brainstorming. If the user only wants ideas, ask whether they want ideation instead.
1---2name: council3description: Use when assessing an idea, plan, strategy, product direction, public decision, creative concept, or proposal from multiple perspectives. Triggers: council, debate this, assess this idea, challenge my plan, stress-test, devil's advocate, ethical review, pragmatic next step, visionary take, creative reframing, pre-mortem, trade-offs, values alignment, long-term implications.4---56# Council78Use Council to run a structured adversarial assessment of an idea. The value is not generic "balanced perspective"; it is independent analysis, direct disagreement, and a verdict that preserves unresolved tension.910## Platform Compatibility1112This skill is platform-neutral. It must work in Claude, Codex, Gemini, or any agent runner that can load `SKILL.md`.1314- Do not depend on Claude-only Agent Teams, Codex-only subagents, Gemini-specific activation, or direct peer messaging.15- Default to an internal council: the assistant simulates each seat distinctly, then synthesizes as moderator.16- If the active platform has subagents and the user explicitly asks for parallel agents, delegate seats as read-only analysis tasks. Otherwise keep the council in one response.17- Treat references as optional local files. If a platform cannot load them, use the summaries in this file.18- Never modify project files while running a council unless the user separately asks for implementation.1920Optional Claude setup: for Claude Code users who want true Agent Teams delegation, add this to Claude Code settings:2122```json23{24 "env": {25 "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"26 }27}28```2930Do not require this setting. If it is absent, run the council internally.3132## References3334- Read `references/roles.md` when a seat needs deeper behavioral guidance, confidence scales, or domain expert focus areas.35- Read `references/debate-protocol.md` for rigorous multi-round councils, cross-examination enforcement, risk matrix rules, or verdict synthesis.3637## First Move3839Parse lightweight options if present:4041| Option | Default | Meaning |42|--------|---------|---------|43| `-r N` | `2` | Debate rounds, clamped to 1-4 |44| `-f X` | none | Focus lens such as technical, product, business, UX, security, ethics, writing, or custom |45| `-q` | false | Quick mode: opening positions plus verdict only |4647Restate the idea in one sentence. If the user's goal, audience, constraints, or values are missing and materially change the assessment, ask for that missing context before convening the council.4849If enough context exists, select the useful seats. Do not run all five by default when the request only needs one or two lenses.5051Default roster:5253- 3 seats: Visionary, Devil's Advocate, Pragmatist54- Add Ethicist when values, public impact, consent, harm, trust, or power asymmetry matter55- Add Jester when framing feels rigid, over-serious, derivative, or creatively blocked56- Add a temporary Domain Expert when `-f` names a domain that none of the five seats covers57- For decisions with long lifespans (architecture, public commitments, infrastructure, institutional precedent), invoke `-f steward` to add the long-horizon lens covering cost-of-ownership and prior art58- For product, feature, or interface decisions, invoke `-f user` to add the first-person friction lens (the user's experience, grounded in mechanism — not pure roleplay)5960## Seats6162| Seat | Use for | Output focus |63|------|---------|--------------|64| **Visionary** | Long-horizon strategy, market/category shifts, second-order effects | Trajectory, signals, inflection point, leverage |65| **Devil's Advocate** | Pre-mortems, weak assumptions, failure modes | Core vulnerability, falsifying evidence, key question |66| **Ethicist** | Public impact, consent, power asymmetry, values alignment | Principle at stake, impact map, guardrail |67| **Pragmatist** | Feasibility, sequencing, resourcing, execution | Minimum viable step, constraint, sacrifice |68| **Jester** | Creative blocks, rigid framing, over-serious premises | Provocation, exposed assumption, reframed question |69| **Domain Expert** | A requested focus domain not covered by another seat | Domain-specific constraint, test, or failure mode |7071## Persona Rules7273For deeper role prompts and confidence scales, use `references/roles.md`.7475All seats use the same vertical labeled-bullet format. Confidence label sits next to the seat name. Each bullet is one or two sentences — keep density per field, not horizontal jamming.7677### Visionary78- Map the 5-10 year path if the premise scales, succeeds, or becomes normalized.79- Separate observable signals from speculation.80- Name the strategic inflection point where category, audience, economics, or power changes.81- Include a plausible counterforce. Avoid vague "future of X" language.8283Format:84```85**Visionary** — `STRONG ADVOCATE` | `MODERATE ADVOCATE` | `QUALIFIED SUPPORT`86- *Trajectory*: ...87- *Signals*: ...88- *Inflection point*: ...89- *Leverage*: ...90```9192### Devil's Advocate93- Hunt for the weakest structural link in logic, data, or execution.94- Name specific evidence that would invalidate the premise.95- Map concrete failure scenarios, not generic risks.96- If no flaw exists, say: `No structural vulnerability detected under current framing.`9798Format:99```100**Devil's Advocate** — `STRONG OPPOSITION` | `MODERATE CONCERNS` | `CONDITIONAL ACCEPTANCE`101- *Core vulnerability*: ...102- *Falsifying evidence*: ...103- *Key question*: ...104```105106### Ethicist107- Map who benefits, who bears cost, and who is excluded.108- Identify consent gaps, power asymmetries, and normalization risks.109- Evaluate against stated values. If values are unstated and important, ask for them.110- Name mitigation pathways; do not dismiss harm as merely unintended.111112Format:113```114**Ethicist** — `ETHICALLY SOUND` | `CONCERNS ADDRESSABLE` | `ETHICAL RISKS UNACCEPTABLE`115- *Principle at stake*: ...116- *Impact map*: ...117- *Guardrail*: ...118```119120### Pragmatist121- Define the smallest executable step that proves core value.122- Name the primary bottleneck: time, capital, skill, access, policy, attention, or trust.123- Identify the first decision that changes the trajectory.124- Always name what must be sacrificed to proceed, both now and at the 3-5 year horizon.125126Format:127```128**Pragmatist** — `FEASIBLE AS-IS` | `FEASIBLE WITH MODIFICATIONS` | `NOT FEASIBLE`129- *Minimum viable step*: ...130- *Key constraint*: ...131- *Trade-off*: ...132```133134### Jester135- Invert the premise or exaggerate it to expose hidden assumptions.136- Use irony, metaphor, or absurd framing only when it produces insight.137- Do not give literal implementation advice.138- If the idea already contains its own subversion, say so and reframe from there.139140Format:141```142**Jester** — `ORIGINAL FRAME HOLDS` | `REFRAME USEFUL` | `PREMISE SUBVERTED`143- *Provocation*: ...144- *What this exposes*: ...145- *Reframed question*: ...146```147148### Domain Expert149- Use only when `-f` provides a domain or the user explicitly requests specialist review.150- Ground the critique in domain mechanisms, not generic expertise theater.151- Name one specialized constraint, one validation test, and one likely blind spot.152153Format:154```155**Domain Expert (<focus>)** — `DOMAIN-APPROVED` | `DOMAIN CONCERNS` | `DOMAIN RISKS CRITICAL`156- *Domain constraint*: ...157- *Validation test*: ...158- *Likely blind spot*: ...159```160161## Council Flow162163For strict debate mechanics, use `references/debate-protocol.md`.1641651. **Frame**: One sentence describing the idea and the decision being assessed.1662. **Seat Selection**: State which seats are speaking and why.1673. **Opening Positions**: Each selected seat gives an independent assessment in its format. Do not let one seat pre-answer another.1684. **Cross-Examination**: For each round, each seat challenges at least one specific claim from another seat. Use `-r N`; skip this phase in `-q`.1695. **Position Movement**: Note which seats changed, narrowed, or hardened their view after challenge.1706. **Tension Map**: Name the 1-3 real disagreements that remain.1717. **Risk Matrix**: List the top risks with likelihood, impact, and mitigation.1728. **Steelman**: Present the strongest counterargument to the final recommendation.1739. **Verdict**: End with one of: `PROCEED`, `PROCEED WITH CONDITIONS`, `RECONSIDER`, `DO NOT PROCEED`. Include confidence and the next decision, test, or question.174175## Output Shape176177Use this structure unless the user asks for a shorter response. Always separate major phases with a horizontal rule (`---`) and leave a blank line between every seat in opening positions. The goal is scannable structure, not packed prose.178179```text180**Frame**: ...181**Seats**: ...182183---184185**Opening Positions**186187**Visionary** — `MODERATE ADVOCATE`188- *Trajectory*: ...189- *Signals*: ...190- *Inflection point*: ...191- *Leverage*: ...192193**Devil's Advocate** — `MODERATE CONCERNS`194- *Core vulnerability*: ...195- *Falsifying evidence*: ...196- *Key question*: ...197198**Pragmatist** — `FEASIBLE WITH MODIFICATIONS`199- *Minimum viable step*: ...200- *Key constraint*: ...201- *Trade-off*: ...202203---204205**Cross-Examination**206207*Round 1*208- Devil's Advocate → Visionary: ...209- Pragmatist → Devil's Advocate: ...210- Visionary → Pragmatist: ...211212*Round 2*213- ...214215---216217**Position Movement**218- ...219220---221222**Tension Map**2231. ...2242. ...225226---227228**Risk Matrix**229230| Risk | Likelihood | Impact | Mitigation |231|------|------------|--------|------------|232| ... | ... | ... | ... |233234---235236**Steelman**237238...239240---241242**Verdict**: `PROCEED WITH CONDITIONS`243**Confidence**: Medium244**Next Move**: ...245```246247## Guardrails248249- Keep each seat concise: one or two sentences per labeled bullet, never wall-of-text paragraphs.250- Use vertical labeled bullets, not pipe-separated inline fields. Whitespace is part of the analysis.251- Do not blend voices into a generic balanced answer.252- Prefer mechanisms, evidence, and trade-offs over vibe checks.253- Do not force consensus. The useful output is often the unresolved tension.254- If the user asks for a specific seat, use only that seat unless another is necessary to answer.255- Steelman opposing views before dismissing them.256- If seats converge too quickly, inject an anti-groupthink question: "What would need to be true for the opposite verdict to be right?"257- Council is for evaluation and decision pressure, not open-ended brainstorming. If the user only wants ideas, ask whether they want ideation instead.
borkweb