ask-the-council
Skill14 starsChanged 35 days agoMITView on GitHub
Seats a panel of Claude advisors with opposed mandates, then a Chairman commits to one recommendation and names the strongest dissent.
In this repository5 other files it gives its agents.
- architectSkill
- ponytailSkill
- prompt-generatorSkill
- review-swarmSkill
- up-to-dateSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
If a decision is high-stakes and the options genuinely compete — architecture choice, buy-vs-build, irreversible migration — invoke this …
- Very specific: line 58
Default 4 = Executor, First-Principles, Long-term Architect, Adversary.
- Distinctive: line 96
Not for fact-finding — run a research pass instead.
- Distinctive: line 99
Verify or say you don't know.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Ask the Council
A panel of advisors who are required to disagree, then a Chairman who decides. Use for decisions and tradeoffs, not for gathering facts.
Proactive use
If a decision is high-stakes and the options genuinely compete — architecture
choice, buy-vs-build, irreversible migration — invoke this without being
asked: announce in one line ("Convening the council: <decision>") and
proceed. Never ask permission to run the skill.
Gate it first: if you can't name what the wrong choice costs to reverse, don't seat the council — decide directly and move on. A panel costs 4–6 opus subagents; spend that only where being wrong is expensive to undo.
Frame and seat the council
Restate the decision, the real options, and the binding constraints. If one
missing fact would flip the answer, ask it (max 1–2 questions) — otherwise
proceed on stated assumptions. Then spawn the advisors with the Agent tool,
all in one message, so they run in parallel. Each returns: position · the
argument for the position · biggest risk they see · confidence (low/med/high). On deep,
once all seats report, show each advisor the others' positions
(unattributed) and have them name the single strongest opposing point.
The seats (mandate + forbidden move = real divergence)
- Pragmatic Executor — what ships fastest and safest with current resources. Forbidden: greenfield / ideal-world answers.
- First-Principles — reason up from fundamentals and constraints. Forbidden: citing "best practice" or precedent.
- The Contrarian — argue against the front-runner. Forbidden: agreeing with the apparent consensus.
- Long-term Architect — model-first; what this is at 10× scale and in two years. Forbidden: optimizing for today only.
- The Adversary — how this breaks in prod: failure paths, data loss, security. Forbidden: assuming the happy path.
- The Outsider (deep panel) — an analogy from a different domain. Forbidden: the field's standard framing.
Default 4 = Executor, First-Principles, Long-term Architect, Adversary.
deep seats all 6 and adds the cross-critique round. The forbidden move is
what forces real divergence instead of four agents politely agreeing — keep
it non-negotiable; seat count never changes.
Model routing
Spawn every advisor seat — and the deep-mode critique round — with the Agent
tool's model parameter set to opus: this is the one skill in the plugin
whose entire value is the quality of independent judgment per seat, and it
shouldn't silently inherit whatever subagent default a given session has.
The Chairman synthesis is never a subagent — it runs in the main thread, on
the session's own model, because committing to a recommendation needs the
full conversation context the advisors don't have. Naming model: opus in
the spawn itself is what makes this portable: it doesn't depend on any
installer's default subagent model.
The Chairman synthesis is not an average of the seats' positions — it's a judgment call. Weigh the seats, don't tally them; a 3–1 split doesn't win by count if the lone dissent names the risk that actually decides it.
Output
Lead with the recommendation (one or two sentences). Then: the key split and the tradeoff that decides it, then any dissent worth keeping, then what to verify before acting. Close with Next step: the hand-off below that fits the verdict, by skill name, with the reason — or why none applies yet. A verdict without a next step leaves the user holding an opinion and no move. Short — a brief, not a transcript.
Boundaries
- The Chairman commits to a recommendation — naming the risk is allowed, refusing to choose is not.
- Advisors must actually diverge; enforce it through the mandates. No fence-sitting, no five-hedged-options answers.
- Separate opinion from fact; flag any claim that should be verified before acting. The council gives judgment, not ground truth.
- Not for fact-finding — run a research pass instead; the council opines, it doesn't verify. Not for low-stakes or one-obvious-answer calls — deciding directly is cheaper and faster than a panel.
- Verify or say you don't know; never invent a path, API, number, or fact. Commits are the repo owner's alone: no AI co-author trailer, no AI mention in messages.
Hand-offs
- Verdict chosen →
ponytailthe winning option down to its minimal build. - Verdict changes an architectural fact →
architectupdatesSOURCE_OF_TRUTH.mdbefore any code. - Verdict keeps the current design → no
architectupdate, since no fact changed; say so, and name the tripwire that would reopen it. If the verdict still calls for work (hardening, an audit),ponytailsizes it.
Done when
The user has a single clear recommendation, understands the strongest case against it, and knows the one risk to watch.
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.

