The Quorum

A decision gets sharper when it survives disagreement.

The Quorum is an agent skill for Claude and Codex that convenes five differentiated experts to pressure-test a decision. Each one forms a verdict independently, then they challenge each other in anonymous peer review, and a chairman synthesizes a calibrated recommendation — with a pre-mortem, a preserved minority report, and a clear next step. It's delivered as an interactive HTML dashboard.

The point isn't consensus. It's surfaced, structured disagreement you can actually act on. Most "get a second opinion" tools quietly optimize to agree with you. This one is built to push back.

---

Why five voices instead of one

Ask one model for advice and you tend to get one smooth, agreeable answer. The Quorum splits that single perspective into five that each own a distinct way decisions go wrong, and forces them to commit before they see each other's reasoning:

Member	Owns	The question they won't stop asking
The Analyst	Epistemics	What do we actually know vs. assume — and what's the one fact that would settle this?
The Skeptic	Downside	What's the strongest case against the leading option?
The Strategist	Time	How does this look in 12 months? 3 years? What compounds?
The Operator	Execution	What does this actually take — and what happens if we just don't?
The Reframer	The frame	Is this even the right question? Is there a third path?

A sixth — The Steward (people, politics, buy-in) — sits on the bench and swaps in for people-heavy decisions like hiring, org design, and partnerships.

Two design choices do most of the work:

• Independence is enforced. Members reason in isolation before peer review. If they anchor on each other, you get five flavors of one answer and the whole thing is theater. Where parallel agents are available they run simultaneously; otherwise each verdict is formed from the original brief alone.
• The Analyst and Reframer aren't forced to pick. The Analyst's job is to name the decisive unknown; the Reframer's is to question the menu. Forcing them to choose from the options would sand off the two voices whose entire value is not choosing from the menu.

---

What you get

An interactive dashboard with:

• The spread — where the members landed and how confident they were, so a clean consensus or a real 3–2 split is obvious at a glance.
• Five member cards — each with a verdict, three bullets, a recommendation (or a decisive unknown / reframe), a confidence level, and the one thing that would flip them.
• Anonymous peer review — relabeled A–E so authority bias doesn't creep in; members who change their mind have to say what specifically changed it.
• A chairman's synthesis — the spread, the real tensions, the call (calibrated to how reversible the decision is), a pre-mortem on the chairman's own recommendation, and a minority report that preserves any high-conviction dissent verbatim instead of averaging it away.

There's also an optional decision journal — log the call and its flip-conditions, then revisit later against what actually happened. Over time it shows you how you decide: where you're well-calibrated, and where you're not.

See it in action: open any worked example in a browser —

• examples/growth-allocation.html — "where should our next dollar of growth come from," showing the council relocate the decision upstream (it's a retention problem, not a growth one) and a high-conviction dissent preserved as a minority report.
• examples/hiring-decision.html — "which finalist do we hire," showing a genuine split, the Steward swapping in, and a minority report.
• examples/side-project-leap.html — "should I quit my job for my side project," showing the council rejecting the framing itself.

---

Install for Claude

1. Download the-quorum.skill from the latest release, then add it through your Skills settings.
2. Add it to Claude via your Skills settings.
3. Convene:

quorum this: should we rebuild our billing system in-house or buy the SaaS tool?

council this works too. You can also just describe a decision — the skill is built to trigger on its own when you're clearly weighing a consequential choice.

Install for Codex

Ask Codex:

Install my Quorum skill from https://github.com/glichtenthal/the-quorum

Or install manually:

python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
  --repo glichtenthal/the-quorum \
  --path . \
  --name the-quorum

Restart Codex after installation.

Good first decisions to try

• Take the offer or stay in your current role?
• Build vs. buy — ship your own version or pay for the existing tool?
• Kill this project / feature, or keep going?
• Hire Candidate A or B for a senior role? (watch the Steward come off the bench)
• Relocate to a new city?

The skill is most useful on decisions with real built-in tension — where smart people would genuinely disagree.

---

What this is, and what it isn't

Worth being straight about:

• The five "members" are one model role-playing five lenses, not five independent intelligences. The value is the structured discipline — independence, committed verdicts, a chairman that checks itself — not literal multiple minds.
• It's a thinking aid, not an oracle. It's at its best clarifying a hard call, not handing you a verdict to outsource your judgment to.
• Don't use it to launder a decision you've already made. If you're hunting for permission, it's built to notice and push back — that's the feature.
• For genuinely irreversible, high-stakes personal decisions — medical, legal, large financial — treat the output as a structured prompt, not a substitute for a qualified professional.

---

Customizing

• Theme: the dashboard ships with a dark palette (--ink #1C1A1D, --lime #E2F075, white text), defined as CSS variables at the top of references/report-template.html. Swap those three values for your own look in one place.
• Roster: edit references/persona-prompts.md to retune a member or change who's on the bench.
• Workflow: SKILL.md is the whole thing in plain language — readable and editable.

Version history

See CHANGELOG.md.

More agent skills

• The Briefing Room - turns messy context into a structured brief a human can think with.
• Ground Truth - calibrated honesty and anti-sycophancy for plans, decisions, reviews, and ideas.
• Test Drive - test an idea, claim, or decision before trusting it.

Build the .skill yourself

The repo is the skill. To package it into an installable file, use the packager from Anthropic's skill-creator tooling:

python -m scripts.package_skill path/to/the-quorum

Repo layout

the-quorum/
├── SKILL.md                      # the workflow
├── agents/
│   └── openai.yaml               # Codex UI metadata
├── assets/
│   ├── quorum-preview.png        # README preview
│   └── social-preview.svg        # README social preview
├── evals/
│   └── evals.json                # behavior checks
├── references/
│   ├── persona-prompts.md        # the five council members + the Steward
│   ├── report-template.html      # the dashboard scaffold (themeable)
│   └── decision-journal.md       # the calibration loop
├── examples/
│   ├── growth-allocation.html    # worked example — council relocates the decision
│   ├── hiring-decision.html      # worked example — a real split
│   └── side-project-leap.html    # worked example — council rejects the framing
├── README.md
└── LICENSE

License

MIT — see LICENSE. Use it, fork it, retune the council, make it yours.