boardroom

Boardroom

Put a decision in front of a boardroom of seven advisors before you make it.

Boardroom is a skill for Claude. You give it a decision; Claude chairs a three-round debate between seven advisor personas, each inspired by a well-known operator or investor's public thinking and each backed by its own researched "brain". The seats score your options, argue with each other, change their minds when they're out-argued, and vote. You get a verdict with conditions, the dissent, and a PDF you can send to your team.

It works in Claude Code, in Claude Cowork, and in regular claude.ai chats.

/boardroom Should we make our first sales hire now, or keep founder-led selling for two more quarters?
Options: A) full-cycle AE now ($120k OTE)  B) founder keeps selling, hire in Q2  C) SDR first ($65k OTE)
Numbers: $41k MRR growing 9%/mo, 45 inbound demos/mo, 31% close, $310k cash, $38k/mo burn
Success looks like: $90k MRR by June with the CEO under 10 hrs/week on sales
I'm leaning: A, because the CEO's time is the bottleneck

Both seats opened for A at 7/10, matching the founder's lean, so the chair treated that as a warning and made them argue for B. They found that the hire adds hours, not demos, and that a 90-day kill switch can't tell a 20% closer from a 31% one. Both moved to B: keep selling, open the search in December, and hire the week a written trigger trips.

A real session, in the repo as a demo.

The seats

Each brain (skills/boardroom/brains/<seat>/) is 11–13k words built from 40–50 public sources:

• a hot sheet: worldview, the eight rules the seat decides by, signature questions, decision loop and known blind spots
• named frameworks
• positions on recurring business topics
• stories
• a voice guide
• the source list

Install

Claude Code (recommended: each seat is a real sub-agent, plus a live UI)

Inside Claude Code:

/plugin marketplace add kaustub1806/boardroom
/plugin install boardroom@boardroom

Then restart Claude Code. If you'd rather not use the plugin system, clone the repo and copy the files in:

git clone https://github.com/kaustub1806/boardroom.git && cd boardroom
./install.sh                 # every project gets /boardroom
./install.sh ~/my-project    # or just one project
./install.sh --uninstall     # remove it again

claude.ai (regular chats)

1. Download boardroom.skill.
2. In claude.ai go to Settings → Capabilities, make sure Code execution and file creation is on, then upload the file under Skills.
3. In any chat, ask something like "put this decision in front of the boardroom". No slash command is needed.

Claude Cowork

Download boardroom.plugin and open it (or drag it in) in Cowork, then confirm the install. The plugin includes the skill and all seven seat agents.

The only dependency is Python 3.9+, which the skill uses to keep the session log and build the PDF.

Use it

Plain language works too: "roast this decision", "run the council on whether to raise prices", "should I do A or B?", "help me write a boardroom prompt".

What to put in your question

The board is only as good as the brief. Give it:

1. The decision, as one question.
2. Options, 2–4 of them, labelled A/B/C.
3. Real numbers: revenue, margin, cash, team, prices, volumes.
4. Limits: deadlines, hours per week, budget.
5. What success looks like, and by when.
6. Which way you're leaning, and why. The room is built to push back on it.
7. Context the board can't guess.

If Claude already has your context (project files, memory, connected tools) it uses that first and only asks about what's missing.

What you get

• In the chat: the decision and its conditions, the vote and weighted scores, the dissent, and the arguments worth rereading.
• verdict.pdf and verdict.md.
• The whole session in ~/.boardroom/sessions/: the brief, each seat's raw reply per round, and an event log the UI replays. Set BOARDROOM_HOME to store it somewhere else.

How it works

you ──▶ chair (Claude)
          │  builds the brief, seats the board
          │  round 1  every seat scores every option, independently
          │  review   unanimity is a warning → two seats must argue for the runner-up
          │  round 2  cross-examination: directed rebuttals, chair pairs the disagreements
          │  round 3  final votes with confidence and conditions
          │  verdict  the chair weighs the arguments (not just the votes), writes it, builds the PDF
          ▼
   boardroom-bezos … boardroom-buffett   (each reads its own brain)

The skill picks one of two modes depending on where it runs:

• Council mode (Claude Code and Cowork). Each seat is its own sub-agent that talks only to the chair, so no seat sees another's draft before round 2.
• Chamber mode (claude.ai chats, which have no sub-agents). The chair plays each seat itself. It opens one brain at a time and writes the seats in a fixed order without revising earlier ones. The rounds, scoring and verdict are the same. You can ask for chamber mode anywhere; it's cheaper.

Scores are weighted by relevance. Each seat rates how squarely the question sits in its lane, from 1 to 3, and that rating weights its scores. So the cash seat's view on cash counts for more than its view on Instagram.

The live room (Claude Code)

python3 ~/.claude/skills/boardroom/scripts/br.py ui --open

A local, zero-dependency viewer:

• the table, with seats that write, speak and score
• arcs showing who is arguing with whom
• the scoreboard and the verdict card, with the PDF download
• past sessions, with replay at 1×, 2× or 4×

To see it before running your own sessions, add --sessions ~/.claude/skills/boardroom/ui/demo-sessions.

Add or swap a seat

/boardroom add <person or role>, for example someone from your industry, or "a CISO". Claude follows ADDING-ADVISORS.md:

1. It defines the seat's lens.
2. It researches 25+ public sources.
3. It writes the brain and the agent file.
4. It registers the seat in advisors.json.

Build from source

./package.sh    # dist/boardroom.skill (claude.ai) and dist/boardroom.plugin (Cowork)

.claude-plugin/         plugin.json, marketplace.json
agents/                 boardroom-<seat>.md: one sub-agent per seat
skills/boardroom/
  SKILL.md              the chair's instructions
  protocol.md           the session and event contract
  advisors.json         the roster
  ADDING-ADVISORS.md
  brains/<seat>/        hot.md, frameworks/, positions.md, stories.md, voice.md, sources.md
  scripts/br.py         sessions, scoring, verdict, PDF, export, UI launcher (stdlib only)
  ui/                   the live room and demo sessions
install.sh, package.sh

Important

• These are AI personas, not the real people. Each seat is inspired by a person's public books, essays, letters, talks and interviews. None of these people are affiliated with this project, and none have endorsed it. The brains are original summaries written in our own words; each lists its sources.
• The verdict is input, not orders. Read the dissent first; it's usually the part that changes your mind.

Credits

Inspired by a video walkthrough of a similar "council" skill. This is an independent implementation.

License

MIT