Claude Council
Convenes 7 expert AI personas to genuinely debate any decision, then delivers a clear verdict with a confidence score — no more "it depends".
SKILL.md is a pure prompt skill: no scripts, no external calls, no data-exfiltration paths, no red-line risks found; least privilege is inherently satisfied. However, there is no disclosure of how sensitive personal inputs (finances, career, relationship details) are handled, no rollback concept, and the publisher is unverified — deducted to 15.
Install and usage paths are clear and internally consistent (upload council/ or paste SKILL.md); the core path is plausible on static reading. But there are no tests, no edge-case handling rules (the author admits in CONTRIBUTING that simple factual questions and grief are open gaps), and weak failure feedback — deducted to 8.
Target scenarios (business/career/technical/financial decisions) and trigger phrases are explicit, and output structure is well defined; but triggers are overly broad (any 'Should I...' may falsely fire), non-fit boundaries are undeclared, there is no Chinese-language support, and the skill depends entirely on claude.ai, limiting mainland-China reachability — deducted to 7.
Documentation is well organized (install, usage, customization, contributing, MIT license, author attribution), but version governance is messy: unpublished v2 staging files exist in the repo, CHANGELOG is not yet created, maintenance responsibility rests on an unverified individual, and referenced example files were not verifiable within the provided sources — deducted to 8.
As a structured decision prompt, the output format (debate, verdict, confidence, risks, next steps, minority report) is complete, and marginal value over a bare prompt is argued via a comparison table; but all example outputs are author-claimed, static review cannot verify direct usability, and the confidence percentages are model self-assessments rather than calibrated measures — deducted to 6 within the static cap.
All evidence is author-claimed example output and marketing material (ChatGPT comparison table, community quotes); there is no third-party execution evidence, no tests, and no independently reproducible material. The 'famous decisions' analyses and example outputs are unauditable — deducted to 3.
- This is a static review only; no code was executed and all example outputs are author-claimed and independently unverified.
- Trigger conditions are overly broad — the Council may fire on ordinary 'Should I...' questions; the stated 'confidence percentage' is model self-assessment, not a calibrated probability.
- The skill depends entirely on claude.ai, an overseas service; mainland-China reachability may be limited and no Chinese-language support is declared.
- The repository contains unpublished v2 staging files; behavior and docs may change. Users disclose personal decision context (finances, relationships) in conversation — mind sensitive-data exposure.
- Publisher identity is unverified; MIT licensing is claimed in repo metadata but not confirmed within the assessed file itself.
What does this skill do, and when should you use it?
Claude Council is a Claude Skill that assembles seven distinct expert personas — the Adversary, Strategist, Scientist, Visionary, Engineer, Philosopher, and Humanist. When you present a decision, idea, or problem, the personas analyze it from their own worldviews, push back on each other, and converge on a structured output: a real debate, a clear verdict with a specific confidence percentage, 3 critical risks, 5 concrete next steps, and a minority report from the strongest dissenter. Each persona has a documented voice, bias, blind spot, and debate behavior. Built by Hussain Ali and MIT-licensed, it suits anyone facing high-stakes decisions who wants genuine disagreement instead of a balanced summary of considerations.
When Claude receives a decision-type question, SKILL.md instructs it to generate analysis for each of the seven persona files under council/personas/ (adversary, strategist, scientist, visionary, engineer, philosopher, humanist), following their documented voices and biases. Output follows templates/ debate-format and verdict-format: a banner-formatted debate, a verdict block with confidence percentage, 3 critical risks, 5 ordered next steps, and a minority report. It is pure prompt structure — no scripts, no external calls.
- A full-time developer considering quitting to build a SaaS, wondering whether to validate first or resign now
- An engineer weighing a job offer that trades $30k salary for 0.4% equity at a Series B startup
- A 4-person startup debating whether to refactor a monolith into microservices
- A founder stress-testing a business plan who wants sharp pushback, not a pros-and-cons list
- Any decision-maker facing real tradeoffs who wants a position taken and an actionable checklist
How do you install this skill?
You need a Claude account (any plan) with the Skills feature enabled in settings. Option 1: git clone https://github.com/itshussainsprojects/Claude-Council-Skill.git, then open Claude.ai → Settings → Skills → Add Skill and upload the council/ folder. Option 2: open council/SKILL.md, copy its contents, and paste into Claude.ai → Settings → Skills → New Skill. INSTALL.md in the repo has a more detailed guide.
How do you use this skill?
After installing, just talk to Claude normally about any decision. Trigger phrases include "Should I...",
How does this skill compare with similar options?
The README contrasts it with "throwing a one-off prompt at ChatGPT": instead of re-pasting a long prompt every time, the skill gives consistent structure, documented biases, real inter-persona debate, calibrated confidence, and a minority report — and it persists. It fits users who want this decision framework installed as a permanent skill.