Decision Council — Development skill for Claude Code

Decision Council

A Claude skill that convenes a **council of independent perspectives** before an important decision: each councilor answers blind from a distinct role (CFO, devil's advocate, operations, brand, growth, risk…), they optionally debate each other, and a final synthesis **takes a stance** instead of averaging.

**The caveat, upfront:** every councilor runs on the same model and shares the same priors. Agreement between them is not independent corroboration — it's a shared starting point to stress-test. The value is in the distinct angles, each role's declared biases, and the blind spots that show up when you cross them. The skill shows this caveat at the end of every session, always.

What it does

• Blind round 1: each member answers in full without seeing the others. Converging out of politeness is banned.
• Optional debate (--debate): a round 2 where each member attacks another's weakest argument, concedes what's genuinely good, and holds or changes their position.
• Deep mode (--deep): per-member self-evaluation + declared confidence + admitted blind spot; the synthesis weights by confidence.
• A synthesis that decides: position map, convergences by different paths, the central divergence with the assumption in conflict, collective blind spots, and a recommendation with a reversal condition. No "it depends".
• 16 roles with personality, guiding questions, declared bias and anti-pattern: 8 business roles (finance, devil's advocate, operations, brand, growth, risk, customer, timing) and 8 technical ones (security, performance, maintainability, simplicity, scalability, dx, compliance…).
• Domain presets: business, sponsor, expansion, product, people, balanced, security-focused, architecture, review.
• Any language: the council replies in whatever language you write in.

Install

**Claude Code**

npx skills add mutantapejack/decision-council

Or manually: clone the repo and copy th