Council Diff

council-diff

5-voice AI council for any decision. Paste a question, get 5 specialist perspectives in parallel, see where they agree and disagree. Brier-audited at resolution. Optional Fable 5 Oracle adjudication for split councils.

When to use this skill

Invoke this skill whenever the user:

• Asks for multiple expert perspectives on a hard call ("what would a VC / lawyer / engineer / spouse think")
• Says the single-model answer feels overconfident or one-sided
• Asks for a "second opinion" or "have a council look at this" or "model council"
• Wants their advice tracked for calibration over time (Brier audit)
• Has a high-stakes founder / engineering / investing / career / product / quant decision

Do NOT invoke for trivial factual questions, single-domain code review (use a code review skill instead), or anything where a single LLM verdict is obviously sufficient.

Setup

npm install council-diff
export ANTHROPIC_API_KEY=sk-ant-...

Usage

import { CouncilDiff } from "council-diff";

const council = new CouncilDiff({ apiKey: process.env.ANTHROPIC_API_KEY });

const result = await council.deliberate({
  domain: "founder",          // founder | engineer | investor | career | product | quant | custom
  decision: "Should I raise a $1M seed or bootstrap?",
  context: "B2B SaaS, $5K MRR, 20% MoM, solo, 12mo runway",
  oracle: "fable-5",          // optional — Mythos-class adjudicator on split councils
  safeMode: false,            // true forces Sonnet 4.6 fallback (zero data retention)
});

console.log(result.recommendation);   // "go" | "wait" | "kill" | "split"
console.log(result.agreement_score);  // 0-1
console.log(result.consensus);        // 1-paragraph synthesis
for (const v of result.voices) {
  console.log(`${v.voice_display} (${v.score}/100): ${v.verdict}`);
}

Built-in domains

• founder — YC Partner / VC Skeptic / Lawyer / Indie CFO / Pragmatic Spouse
• engineer — Rust Maintainer / SRE Oncall / Recruiter / Junior Dev / CTO 5y Later
• investor — Macro / Sector / PM / Growth VC / Activist Short
• career — Mentor 20y / Recruiter / Peer Doing Well / CSO / Future You 5y
• product — Real User / Competitor / Internal Dev / Garry-style / Naval-style
• quant — Jane Street MD / Citadel / Two Sigma ML / Anthropic / HFT Engineer
• custom — fully user-defined voice rosters

Brier audit

council.audit() reads ~/.council-diff/forecasts.jsonl and scores each voice + Oracle against ground truth at resolution. Tracks per-voice calibration over 30 / 90 days so you learn which voice tends to be right in which domain.

Privacy

Pass safeMode: true to force every Mythos-class request down to Sonnet 4.6 (zero data retention). The Oracle response includes data_retention: "30day-mythos" | "zero" so apps with privacy claims (mental-health, GDPR-sensitive PII, on-device-only) can branch on it.

License

MIT. Source: https://github.com/alex-jb/council-diff

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: alex-jb
• Source: alex-jb/council-diff
• License: MIT
• Homepage: https://www.vibexforge.com/council

Install and usage instructions live in the source repository linked above.