Debate Council — an adversarial debate council for Claude, plus a Boardroom that executes its rulings
Version 1.2.0
A custom Claude Skill that stress-tests a question or decision by arguing it against itself before answering.
When you summon it, Claude convenes ten debaters — five for the motion, five against, each assigned a different angle (financial, risk, execution, long-term, values, base rates, opportunity cost, second-order effects) — runs a parliamentary-style debate with constructives, rebuttals and closings, then a neutral judge scores every surviving argument against a published rubric and issues a ruling.
The judge is not allowed to flatter you. It records its initial lean before hearing arguments, must name the strongest surviving objection to its own verdict, and may rule For, Against, Conditional, or Insufficient evidence. By default you get one line — a 25–30 word verdict — and nothing else. The full debate is one word away.
The second skill, Boardroom, picks up where the council stops: it takes the ruling and decides whether it can actually be executed.
Both skills answer only to words you pick. There is no imposed name — call them what you like in your own language. Defaults are debate council and boardroom; change either in one step (below). Until you do, each skill reminds you once per conversation that you can.
The skill fires only when your message contains your summon word. Ordinary questions, "pros and cons", and debate-adjacent language do nothing.
| You write | Claude runs |
|---|---|
<word> <question> |
Full debate internally; one-line verdict (25–30 words), nothing else |
<word> full <question> |
Complete transcript: all ten constructives, cross-rebuttals, closings, scorecard, ruling |
<word> gameplan <problem> |
Gameplan mode: the top 3 research-backed fixes, each citing a real, verified source with an evidence tier |
gameplan works standalone on any problem, or immediately after a ruling — in which case it targets the points the ruling exposed.
- Motion — your question rewritten as a single debatable proposition
- Judge's prior — the judge's initial lean, written before any argument is heard
- For / Against — the three strongest arguments per side, each tagged with its angle and the fact it rests on
- Rebuttals that landed — which points survived cross-examination
- Verdict with confidence, the decisive factor, a mandatory Dissent, a Prior check (Held / Moved), and What would flip it
- Facts first. If the motion turns on verifiable facts, Claude searches before debating. Debaters cannot cite numbers that don't exist; a fabricated statistic strikes the argument.
- Steelman only. Each bench must make the case its most competent advocate would make. If the judge cannot name a surviving objection to its verdict, the losing side was strawmanned and the debate is rerun.
- Scored, not counted. Every argument gets an evidence tier (verified data → peer-reviewed → professional standard → practitioner → anecdote), a validity score and a rebuttal-survival score. Weight is the product, so one decisive argument beats three weak ones. See
debate-council/references/rubric.md. - Ambiguity is a question, not a guess. If a material constraint is missing, Claude asks 1–3 precise questions before convening.
- Scope boundary. Medical, legal and high-stakes financial motions are labelled as structured analysis, not professional advice. The debate still runs.
- Gameplans cite only what was actually found. Every fix names its technique and a source Claude verified in-session, with an honest evidence-strength label.
The council decides what. The Boardroom decides whether it can be executed and makes the first move. It never re-argues the merits.
| You write | Claude runs |
|---|---|
<boardroom-word> |
Picks up the latest council verdict (or a decision you state), runs the full session, outputs 1–2 lines: GO / GO-IF / NO-GO + first move |
<boardroom-word> full |
Complete minutes: pre-mortem by officer, advisory citations, positions, 30/60/90 plan, risk register, kill criteria |
Who sits at the table. CEO chairs. CFO, CTO, COO, Chief Risk/General Counsel and CHRO each argue only from their seat and only with numbers. VP-level owners supply effort per workstream. Three non-voting professors (strategy, finance, organisational behaviour) test every argument against a named framework — Porter, Rumelt, Kotter, McKinsey 7S, PMBOK, Theory of Constraints, real options, Klein's pre-mortem, COSO and others listed in boardroom/references/roles-and-frameworks.md. A misapplied or invented framework is struck, same as a fabricated figure.
How it rules. Intake → real numbers (searches for external ones, labels internal estimates as estimates) → pre-mortem → each officer states GO / GO-IF / NO-GO → CEO rules and assigns one owner, one action, one deadline.
The guardrail. On NO-GO the committee stops and hands you three choices: override and execute, send it back to the council with the committee's evidence, or drop it. It never softens a NO-GO into a GO-IF to avoid the confrontation and never routes to the council on its own. If an officer's real objection is "the decision is wrong," the output says so — that is council territory.
Requires a Claude Pro, Max, Team or Enterprise plan with code execution enabled. Free plans can't install custom skills.
Option A — pick your word, then install (recommended)
git clone https://github.com/duhan023/Debate-Council-System.git
cd Debate-Council-System
python setup.py --council-word "your word" --boardroom-word "your other word"This rewrites both skills with your summon words and produces releases/debate-council.skill and releases/boardroom.skill on your machine. In Claude, open Settings → Capabilities, turn on Skills, and upload both files. Either flag can be omitted to keep that skill's default.
No Python? Open debate-council/SKILL.md (and/or boardroom/SKILL.md) in any text editor, find-and-replace the default word with yours (replace all), zip each skill folder, rename the zip to <folder>.skill, and upload.
Option B — install as-is, no terminal
Click the green Code button → Download ZIP, unzip it, then zip the debate-council folder and the boardroom folder separately and rename the zips to debate-council.skill and boardroom.skill. Upload both in Claude. Summon with debate council and boardroom. You can rename later with Option A and re-upload.
Claude Code — copy the debate-council/ and boardroom/ folders into ~/.claude/skills/ (personal) or .claude/skills/ inside a project. Run setup.py first if you want your own word.
debate-council/
├── SKILL.md # the council: summon word, modes, steps, output formats, judge's standard
└── references/
└── rubric.md # scoring rubric the judge applies before ruling
boardroom/
├── SKILL.md # the committee: hierarchy, pre-mortem, GO / GO-IF / NO-GO, first move
└── references/
└── roles-and-frameworks.md # officer accountabilities and the frameworks the advisory bench may cite
setup.py # set your summon words and package both skills into releases/
Issues and pull requests are welcome. Keep SKILL.md under 500 lines; put anything longer in references/. Changes to the rubric should come with a short note on why the weighting moved.
MIT — see LICENSE.