council
Skill0 starsChanged 5 months agoMITView on GitHub
Convene a read-only multi-agent review council to stress-test a genuine judgment call before committing — go/no-go, strategy, a hard trade-off, "should I commit to X?". Assembles a best-fit 4–6 seat panel under hard adversary + reality floors, runs a two-round blind→adversarial debate, and produces an auditable, dissent-preserving verdict. Invoke as /council [preset] [quick|standard] [+specialist…] <idea>. SKIP FOR trivial decisions with an obvious answer, factual lookups (just answer), code review (use /code-review), and anything already settled.
167 other projects use a version of this.
Adopted by
Most recently taken up by
- PaulWoitaschek/Voicetheir own version, first seen Nov 16, 2026
- syanga/astacktheir own version, first seen Nov 16, 2026
- JorgeMenaDev/skillstheir own version, first seen Nov 16, 2026
- rodeyseijkens/skillstheir own version, first seen Nov 16, 2026
- Edzard9/stashtheir own version, first seen Nov 16, 2026
- chrisezedi/claude-skillstheir own version, first seen Nov 16, 2026
Used with
Projects with it also use agent-payment-x402, motion-advanced and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
TironFoxSkillagent-payment-x402
Add x402 payment execution to AI agents — per-task budgets, spending controls, and non-custodial wallets via MCP tools. Use when agents need to pay for APIs, services, or other agents.
Also in 5 projects that use this
renepfSkillmotion-advanced
Advanced motion patterns for React / Next.js — drag & drop, gestures, text animations, SVG path drawing, custom hooks, imperative sequences (useAnimate), loaders, and the full API decision tree. Requires motion-foundations.
Also in 4 projects that use this
RojiosSkilldefi-amm-security
Security checklist for Solidity AMM contracts, liquidity pools, and swap flows. Covers reentrancy, CEI ordering, donation or inflation attacks, oracle manipulation, slippage, admin controls, and integer math.
Also in 6 projects that use this
scottwaterSkillmodernize
Rebase an aging PR onto its current base and decide whether it should be modernized, reworked, or closed. Use when asked to rebase, update, or modernize an old PR.
Also in 7 projects that use this
RojiosSkilldart-flutter-patterns
Production-ready Dart and Flutter patterns covering null safety, immutable state, async composition, widget architecture, popular state management frameworks (BLoC, Riverpod, Provider), GoRouter navigation, Dio networking, Freezed code generation, and clean architecture.
Also in 5 projects that use this
scottwaterSkillleaving-pr-comments
Format comments an agent posts on GitHub on the user's behalf. Use whenever leaving a PR comment, review comment, or reply.
Also in 6 projects that use this
- +1 more from scottwater
- +1 more from Rojios
Other versions167 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 167
- simstudioai/sim★ 30ksame name
- 0xNyk/council-of-high-intelligence★ 4.9ksame name
- t3dotgg/tstack★ 3.8ksame name
- PaulWoitaschek/Voice★ 3.3ksame name
- danieleteti/delphimvcframework★ 1.4ksame name
- warpdotdev/common-skills★ 630same name
- QuixiAI/agi-memory★ 623same name
- QuixiAI/Hexis★ 623same name
- zaxbysauce/opencode-swarm★ 532same name
- ZaxbyHub/opencode-swarm★ 531same name
157 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository13 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- audit-extremeSkill
- auditSkill
- capture-noteSkill
- capture-taskSkill
- daily-catchupSkill
- dailySkill
- dreamSkill
- project-scaffoldSkill
- projectSkill
- setupSkill
- weeklySkill
- workspaceSkill
The file
We split this skill into the rules it gives an agent
12 rules. 0 are common, 9 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 150 edited versions we read keep these.
9
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 76
If a preset was named → use that fixed lineup (above).
- Distinctive: line 84
If a draft lineup misses a floor, add the missing seat (dropping the least-relevant other seat to stay at ~5).
- Very specific: line 87
Add any explicitly requested with +.
- Distinctive: line 127
Adaptive stop: if round 1 is already strongly unanimous, OR cleanly divergent on a single axis, skip round 2 and go straight to synthesis — …
- Distinctive: line 150
Run the synthesis at the council's tier model.
- Very specific: line 170
When writing a _meta/council/ record, fill related: with at least one [[wikilink]] to a related note (link hygiene), and set council:/tier: …
- Very specific: line 177
Read-only agents: Every seat gets only Read + Grep.
- Distinctive: line 178
Only this skill (main thread) writes, and only the ONE verdict record, and only after explicit confirmation.
- Distinctive: line 182
Selection by coverage + tension, not relevance: Do not assemble a sympathetic panel.
- Distinctive: line 183
Every council satisfies the adversary floor AND the reality floor.
- Distinctive: line 189
Two rounds, resist a third: Use the adaptive stop; never run a round 3.
- Distinctive: line 192
Degrade gracefully: If a subagent dispatch fails, synthesize from the seats that returned and state which were missing.
“Common” is measured: kept by at least 6 in 10 of the 150 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Council Skill
What this does
Convenes a small panel of read-only persona agents to stress-test an idea or decision before the user commits. It assembles a best-fit 4–6 seat council under hard floors, runs a two-round debate (blind independent takes → anonymized adversarial reaction), then synthesizes a structured verdict as a chair. The verdict is shown in chat and, only on confirmation, written to a dated record. The entire mechanism exists to defeat one failure mode — performative agreement / bland consensus — so it forces genuine, legible disagreement at every step.
This skill is decision support, never an oracle. It does not auto-act on its recommendation and does not write without explicit confirmation.
Invocation
/council [preset] [quick|standard] [+specialist…] <idea or question>/council <idea>(default) → assemble a best-fit 5 under the §floors below, print the panel + per-seat rationale, and get a one-tap confirm (or swap/override)./council <preset> <idea>→ force a named preset (below), skipping fluid selection. Presets:general,venture,technical(a.k.a.security),academic(a.k.a.life),creative.+specialisttokens → append a specialist:+architect,+cybersec,+legal,+user-advocate,+finance. The skill also AUTO-SUGGESTS specialists from the idea.quick/standard→ override the inferred model tier (standard= Opus, default;quick= Sonnet).
The roster
Fourteen persona agents in .claude/agents/ (each read-only, Read+Grep). Nine core
lenses + five specialists (✦):
| Seat | Lens |
skeptic | Attacks the idea as-stated; present-tense objections. (adversary) |
pre-mortem | Assumes it already failed a year out; works backward. (adversary) |
pragmatist | Smallest test that proves/kills it; execution reality. (reality) |
resource-realist | The user's personal time/money/attention; opportunity cost. (reality) |
historian | Reads vault decision history; "you faced this before, chose X." |
first-principles | Strips assumptions, rebuilds; "why exist at all?" |
market | Demand test; would anyone use/pay? Carries the upside-as-evidence. |
north-star | Alignment with the user's real stated goals. |
consequences | Second-order effects; third parties; "and then what?" |
architect ✦ | Technical soundness, maintainability, design integrity. |
cybersec ✦ | Threat model, attack surface, abuse cases (hold a high bar if security is the user's domain). |
legal ✦ | Contract/liability/regulatory/IP risk. NOT legal advice. |
user-advocate ✦ | The end user's lived experience (e.g. a customer, a reader). |
finance ✦ | A venture or project's unit economics: pricing, margins, runway. |
Named presets (shortcuts / fluid-selection priors):
| Preset | Members |
general (default fallback) | skeptic · pragmatist · historian · pre-mortem · north-star |
venture | market · pragmatist · resource-realist · pre-mortem · consequences |
technical / security | architect · cybersec · skeptic · pragmatist |
academic / life | historian · north-star · resource-realist · skeptic · consequences |
creative | first-principles · skeptic · pragmatist · pre-mortem |
Steps
Phase 0 — Setup (this main thread)
- Parse the invocation: pull out an optional preset name, an optional
quick|standardtier, any+specialisttokens, and the remaining text as the idea/question. - Compose the council:
- If a preset was named → use that fixed lineup (above).
- Else assemble a best-fit 5 by coverage + tension, NOT relevance: pick the lenses most likely to challenge and expose blind spots in this idea — explicitly NOT the ones that fit its framing most comfortably. Start from the closest preset as a prior. An idea described with enthusiasm must still get its hard seats.
- Enforce the floors (non-negotiable):
- Adversary floor: ≥1 of
skeptic/pre-mortem. - Reality floor: ≥1 of
pragmatist/resource-realist. If a draft lineup misses a floor, add the missing seat (dropping the least-relevant other seat to stay at ~5).
- Adversary floor: ≥1 of
- Specialists (§4.2):
- Add any explicitly requested with
+. - Auto-suggest by scanning the idea for triggers and proposing before running:
"call / text / phone customers" →
+legal +user-advocate; "handle payments / pricing / subscription" →+finance; "store / collect user data" →+cybersec +legal; architecture / system-design language →+architect. The user confirms or declines. - Stay in the sweet spot: the base council is 5. Adding specialists may reach 6. At ≥6 seats, either note the added cost OR swap out the least-relevant base seat rather than always appending — keep the panel from bloating past the point where synthesis turns to mush.
- Add any explicitly requested with
- Resolve the model tier (§3.1): infer from stakes — real/irreversible/high-cost decision → Standard (Opus) (default); low-stakes gut-check / fast iteration → offer Quick (Sonnet). Never Haiku for any seat (weak-link effect). State your pick and note that a Standard run is ~11 Opus calls (~5 seats × 2 rounds + chair) and burns the usage window faster than Quick.
- Print the proposed panel — each seat with a one-line rationale — plus the tier, and CONFIRM. The user may confirm, swap a seat, override the whole lineup, or change the tier before round 1 runs.
- Build the context brief ONCE (bundled to every seat so they don't each re-read the
vault): the idea/question;
_meta/orient.mdalways;_meta/profile.md, the relevant projectREADME.md, and prior_meta/council/records as relevant. (Thehistorianseat additionally reads decision history itself — that's its job.) - Critical anchoring guard: the brief contains the idea and relevant facts but NOT the user's lean. Never tell agents "I'm leaning yes / I want to do this" — it pulls them toward agreement and destroys the council's value.
Phase 1 — Round 1: blind independent takes (parallel)
- Dispatch each council member as a read-only subagent via the
Agenttool, in parallel (one message, multipleAgentcalls), each with isolated context (no conversation history, no other seat's output). Usesubagent_type: <seat-slug>and set the per-callmodeloverride per the tier: Standard →model: opus; Quick →model: sonnet. This independence is the primary groupthink defense. - Pass each seat: the context brief + the idea. The seat returns its
## Round 1output contract (Stance / ≥2 objections / top reasons / one sharp question / confidence) — the contract is already baked into the agent file.
Phase 2 — Round 2: adversarial reaction (parallel)
- Anonymize the round-1 takes as "Response A / B / C…" (strip which persona said what) and identify the 1–2 sharpest points of disagreement.
- Adaptive stop: if round 1 is already strongly unanimous, OR cleanly divergent on a single axis, skip round 2 and go straight to synthesis — extra rounds entrench rather than improve. Otherwise continue.
- Re-dispatch each seat (subagents are stateless — re-inject its own round-1 take + the anonymized others) with an adversarial instruction: "Steelman the strongest opposing view among these anonymized responses, then defend, concede, or refine your own position. Find the strongest objection to YOUR view." Framing is never "move toward agreement." Each returns a final stance + what (if anything) changed its mind. Same tier model as round 1.
Phase 3 — Chair synthesis (this main thread — treat yourself as a biased judge)
- Synthesize the verdict with these guards (you are an LLM-as-judge and inherit its
biases — these counter them):
- Integrate, don't tally — weave the strongest points into a directional recommendation; do not just count Support/Oppose votes.
- Randomize the order you consider takes (position-bias guard); attribute each point to the seat that raised it (verbosity / self-preference guard).
- Rank objections by likelihood × impact — no false balance; not every objection is equal.
- Preserve dissent — the minority view and which seat held it appears in the verdict; never average it away.
- The
legalseat's contributions carry its not-a-lawyer disclaimer into the verdict. - Run the synthesis at the council's tier model.
Phase 4 — Verdict output
- Show the verdict in chat, structured per
templates/council-verdict.md. Read that template and substitute every placeholder:{{date}}(now, YYYY-MM-DD);{{council-label}}(preset name or "custom");{{tier}}(Standard/Quick + model);{{question-oneline}}/{{question}};{{members}}/{{council-lineup}}(seats + one-line rationale each);{{rounds}}(1 or 2, noting any adaptive stop);{{recommendation}}(a directional call, not a shrug);{{case-for}}/{{case-against}};{{objections-ranked}}(by likelihood × impact);{{risks}};{{conditions}}(falsifiable "what would have to be true");{{open-questions}};{{confidence}}(High/Med/Low);{{human-decisions}}(what only the user can decide);{{dissent}}(minority view + which seat). Any empty section →—. - Offer to persist (ASK FIRST — never auto-write):
- Advisory deliberation →
_meta/council/YYYY-MM-DD-<slug>.md. - Committed vault-wide decision → an ADR appended to
_meta/decisions.md(inline, that file's format). - Committed project decision → an ADR at
10-projects/<project>/decisions/usingtemplates/decision-log-adr.md. When writing a_meta/council/record, fillrelated:with at least one[[wikilink]]to a related note (link hygiene), and setcouncil:/tier:in frontmatter.
- Advisory deliberation →
- Operational follow-ups (tasks, reminders) are handed to
/capture-task— never invented in the vault (the operational/durable bridge).
Rules
- Read-only agents. Every seat gets only
Read+Grep; they cannot edit or write. Only this skill (main thread) writes, and only the ONE verdict record, and only after explicit confirmation. - Never auto-persist. The verdict write is opt-in per run.
- Anchoring guard. Agents never receive the user's stated lean.
- Selection by coverage + tension, not relevance. Do not assemble a sympathetic panel. Every council satisfies the adversary floor AND the reality floor.
- No fabricated authority. Seats are abstract evaluative frames, not impersonations of real named experts.
legalis not legal advice. It flags risk and names what to verify with a real attorney; its verdict contributions carry that disclaimer; it never states a definitive legal conclusion.- Two rounds, resist a third. Use the adaptive stop; never run a round 3.
- No Haiku for any critique seat (weak-link effect); model choice is capability/cost only, never a diversity lever — diversity comes from the personas + the mechanics.
- Degrade gracefully. If a subagent dispatch fails, synthesize from the seats that returned and state which were missing. If a tool/MCP is unavailable, say so rather than guessing.
- Operational follow-ups go to
/capture-task, not the vault. - Wikilinks, not markdown links, in any vault note this skill writes.
- Does not edit
_meta/profile.mdor_meta/memory.md.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

