What does this skill do, and when should you use it?
Claude Council is a Claude Skill that, once installed, analyzes any decision or idea through 7 expert personas: the Adversary, Strategist, Scientist, Visionary, Engineer, Philosopher, and Humanist. Each persona argues from its own lens and responds to the others in a single debate round, followed by a structured verdict with a clear position, a specific confidence percentage, exactly 3 critical risks, and 5 concrete next steps. Each persona's voice, bias, and blind spot live in separate files, and the skill calibrates which personas dominate based on input type (startup, career, technical architecture, ethical dilemma, etc.). The repo is MIT-licensed and developed by Hussain Ali.
When triggered, the skill first decides whether the input merits a council (simple factual questions don't qualify); it may ask one clarifying question if critical context is missing. It then reads the 7 persona files plus 2 format templates, generates one round of persona statements with cross-referencing debate per debate-format.md, and produces a verdict block per verdict-format.md: position, confidence percentage with rationale, exactly 3 critical risks, 5 next steps, and a minority report from the strongest dissent. It calls no scripts, network, or external tools — it is pure prompt engineering.
- An employee weighing quitting their job to go full-time on a side project, who wants to be challenged from multiple angles before committing
- An engineer making a solo technical architecture call who needs to know where the system breaks at scale
- A candidate evaluating a job offer with equity who needs dilution and real costs spelled out
- A founder with a startup idea who wants the fatal flaws surfaced
- Anyone facing an ethical or personal dilemma who wants first-principles analysis
- A creative professional stress-testing a project before investing months in it
How do you install this skill?
- No Chinese-language support declared; English trigger phrases may fire unreliably for Chinese users.
- Confidence percentages and base rates are model-improvised and may be precise-looking but ungrounded — do not treat them as facts.
- Triggers are extremely broad (any described plan can invoke), creating false-trigger interference with other tasks.
- No versioning or changelog; future changes are not traceable.
- Output is structured opinion, not verified conclusions; key facts still need human verification for high-stakes decisions.
- Local filesystem
1) Clone the repo: git clone https://github.com/itshussainsprojects/Claude-Council-Skill.git; 2) Open Claude.ai → Settings → Skills → Add Skill; 3) Upload the council/ folder (containing SKILL.md, personas/, and templates/); 4) Start talking — the Council triggers automatically. The repo's INSTALL.md mentions 3 installation methods, but their details are not reproduced in the source.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/itshussainsprojects/Claude-Council-Skill.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/claude-council-skill/council" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Should I quit my job to go full-time on my SaaS?
- Help me decide
- Convene the Council
Just ask with a trigger phrase, e.g. "Should I quit my job to go full-time on my SaaS?", "Help me decide", "Stress-test this", "Convene the Council" — or simply describe a plan or dilemma and the skill fires on its own. After a session, follow-up questions are answered conversationally; the full Council only re-runs if you explicitly ask.
What are this skill's strengths and limitations?
- Strict output structure: clear position, confidence percentage, exactly 3 risks, 5 action items — no hedging
- 7 personas with documented voices, biases, blind spots and signature phrases; the debate genuinely cross-references, not parallel monologues
- Automatically calibrates persona prominence by input type (e.g. startup inputs foreground Adversary/Strategist/Humanist)
- Verdict includes a minority report preserving the strongest counterargument
- MIT-licensed, free, and persona files are editable and extensible
- Not suited to simple factual questions or quick how-to tasks — only judgment calls with tradeoffs
- Quality rests entirely on prompts — no code, no test suite, and cross-model consistency is unverified
- GitHub metadata lists license as unknown; only the README badge claims MIT, so verify independently
- Output is long (7 persona statements plus a verdict) — overkill for quick questions
- Confidence percentages and 'base rates' are model-generated and may look precise without verifiable sources
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Claude Council this page | 57 · Use with care | ★ 97 | 3mo ago | — |
| Claude Council | 47 · Use with care | ★ 97 | 3mo ago | — |
| Justin Sun Cognitive OS (Sun Skill) | 47 · Use with care | ★ 338 | 6mo ago | — |
| Healthcare Agents — 51 US Healthcare Admin Specialists | 59 · Recommended | ★ 52 | 21d ago | Apache-2.0 |
| Ten-Principle Thinking Loop | 53 · Use with care | ★ 16k | 1mo ago | MIT |
The README positions it against 'writing a 2000-word ChatGPT prompt': the skill is essentially that prompt, already written, tested, and free, with consistent structure, documented persona biases, and a minority report — without re-pasting each time.
How did FollowSkills review this skill?
Pure prompt-based skill: no scripts, no external calls, no file writes or network requests; minimal privilege surface. Data flow is transparent (only reads local persona/template files). Deductions: publisher identity unverified; the skill auto-reads 7 persona files and activates on broad trigger phrases, with no explicit user confirmation of convening.
Referenced files (personas/*.md, templates/*.md) are present and instructions are self-consistent; output format rules are detailed. Deductions: static review cannot execute; no tests, no error-handling guidance for abnormal inputs (empty, overly long, non-decision inputs); overly broad triggers risk false invocation.
Scenarios are clear; 'When to Convene / Do NOT convene' provides positive and negative boundaries plus a per-input-type persona calibration table. Deductions: trigger phrases are very broad (any described plan triggers), creating clear false-trigger risk; no Chinese-language support; reachability from mainland China depends entirely on Claude itself.
Well-layered docs (SKILL.md + personas + templates), README with install steps, examples, contributing guide, MIT license. Deductions: no version number or changelog, no known-limitations disclosure, maintenance responsibility and update path unclear, publisher unverified.
README includes 3 full example outputs in directly usable format; clear marginal value over hand-writing prompts. Deductions: examples are author-generated with no third-party execution evidence (static cap is 7); the 'specific confidence percentages and base rates' are model-improvised and may be precision without grounding, requiring user review.
Example files and a demo site provide some auditable material. Deductions: all evidence is self-authored; example outputs are not independently reproducible; no test suite or CI evidence covering the skill's key paths; community quotes are marketing-style.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →