English · 简体中文 · Try it in your browser

The Pentagon doesn't send a general to testify before Congress without a murder board: a panel whose only job is to ask the most brutal questions first. NASA runs them before flight reviews. PhD advisors run them before defenses.

Now you get one before Monday's meeting. It reads your deck, works out the questions most likely to sink you, grills you one question at a time, scores every answer, and hands you a prep sheet.

Animated demo: the murder board reads a deck, seats a panel, asks the CEO's hardest question, scores the answer and suggests a better one

Install

Claude Code, Codex, Cursor, Gemini CLI, Copilot, OpenCode, OpenClaw and 40+ other agents

npx skills add MbappeWU/murder-board

Or paste this into your agent:

Install the murder-board skills from https://github.com/MbappeWU/murder-board (follow INSTALL.md)

Claude Code plugin · Claude.ai · ChatGPT, Gemini, DeepSeek, Kimi and other chat apps

Claude Code plugin marketplace

/plugin marketplace add MbappeWU/murder-board
/plugin install murder-board@murder-board

Claude.ai: download murder-board.zip from the latest release and upload it in Settings → Capabilities → Skills.

Any chat AI (no install): use the web page, or copy a portable prompt into the chat, a Project, a custom GPT or a Gem:
murder-board · pre-mortem · boardroom · so-what · sparring · 中文版

Manual: copy the folders in skills/ into your agent's skills directory (see INSTALL.md).

Then just ask:

murder-board: grill me before Thursday's exec review. Deck: ./q3-plan.pdf

How it works

	Step	What happens
🔎	Recon	Reads your deck, doc, PRD or resume like an opponent: the ask, numbers that don't reconcile across slides, the assumptions, what's missing, and what the room already believes.
🪑	The panel	Seats 3–5 people who would really be in that room (a CFO, a skeptic, your boss), each with an agenda and a signature move. Describe your real boss and they get a seat.
🎯	The kill list	Drafts 15–25 questions ranked by Kill Score = likelihood × damage. You don't see it until the end, just like the real thing.
🔥	The drill	One question at a time, in character. Dodge and it follows up ("That's not what I asked."). Ramble on Brutal and it interrupts.
📋	Score cards	Every answer scored on Directness · Evidence · Honesty · Brevity · Composure, with a "Say this instead" rewrite in your own voice.
🧾	Prep sheet	A readiness score, the 3 landmines, 10 must-nail answers, what to bring, what to cut from your deck, your first 30 seconds and your bad habits.

Three difficulties: Friendly, Realistic, Brutal. Say pause at any time to step out and get coaching. Short on time? Quick mode gives you the top-10 kill list with model answers in one shot.

📖 Read a full example session →

Presets

Pick one, or just describe the room.

Moment	Preset	Default panel
Leadership review	`exec-review`	Boss · CFO · Operator · Skeptic
Budget or headcount ask	`budget-ask`	CFO · Boss · Operator · Peer stakeholder
Justifying your team's efficiency	`efficiency-review`	Skeptical exec · HR · CFO · Technical expert
Board meeting or QBR	`board-meeting`	Investor · Independent director · CFO · Operator
Investor pitch	`investor-pitch`	Lead partner · Skeptical partner · Operator partner
Job interview	`job-interview`	Hiring manager · Bar raiser · Future peer
Promotion committee	`promo-committee`	Chair · Skip-level · Skeptic · HR
Thesis or PhD defense	`thesis-defense`	Chair · Methods expert · Outside member
Design or architecture review	`design-review`	Principal engineer · On-call owner · Security
Outage postmortem	`incident-review`	Eng leader · SRE · Customer-facing lead · Peer lead
PRD or launch review	`product-review`	Product lead · Customer · Engineering · Data
Sales pitch	`sales-pitch`	Economic buyer · Skeptical boss · End user · Procurement
All-hands Q&A	`all-hands`	Two employees · HR · Skeptic
Press or crisis	`press`	Journalist · Skeptic · Legal
述职 · 竞聘 · 立项评审 · 课题申报答辩 · 结构化面试 · 考研复试	Chinese presets	Run entirely in Chinese

The rest of the kit

Four companion skills with the same philosophy: be hard on the plan now so reality doesn't have to be.

Skill	Use it when	What you get
pre-mortem	You're about to commit to a plan	"It's six months later and it failed. Why?" Failure stories from every angle, early-warning signals, tripwires, and the assumptions to test this week. A team facilitation kit too.
boardroom	You face a hard decision	A personal board of five directors who disagree with each other and with you. Independent memos, cross-examination, a vote with dissent, and the Chair's verdict.
so-what	You're writing for busy executives	Your update, email or report rebuilt answer-first (Pyramid Principle), in 3 lengths, with a red-pen report of what was cut and why.
sparring	A hard 1:1 conversation is coming	The AI plays the other person (raise, feedback, layoff, negotiation, resignation), reacts to exactly what you say, then scores you and writes your script.

Why it works

It won't flatter you. Assistants are trained to agree with you. The room won't. These skills are built to disagree first.
Specific, not theatrical. It quotes your numbers and your slides. "Slide 7 says 40% margin, slide 12 implies 22%. Which is it?" beats generic toughness.
It audits the document too. Contradictory numbers, promised-but-missing data, and footnotes that quietly undercut your own case.
It never invents your facts. Model answers use [placeholders] for anything you haven't told it.
Your language, your room. Speak Chinese and the whole board speaks Chinese, with presets for 述职, 竞聘 and 立项评审.
Just Markdown. No server, no API key, no telemetry. It runs inside the AI you already use.

FAQ

Isn't this just a prompt?

It's a structured workflow in plain Markdown: recon, panel design, Kill Score ranking, a drill protocol, a calibrated rubric and a prep-sheet format, plus libraries of panel archetypes and scenario presets. Read it all in skills/murder-board. That it's plain text is a feature: you can read it, fork it and tune it for your own company.

Is it too harsh?

Pick Friendly for a first run, or if you're nervous. The rules forbid personal attacks: pressure goes on the argument, never on you. Say pause at any moment to step out of the room.

Which model should I use?

Any capable model works. Stronger reasoning models produce sharper kill lists. Voice mode (ChatGPT, Claude, Gemini, Doubao) makes the drill feel like the real room.

Is my deck sent anywhere?

Not by this repo. The skills are text files that run inside your own AI tool, under that tool's privacy terms.

Contributing

The best additions are presets (who sits on the panel for your high-stakes moment) and killer questions (the question that actually sank someone). Open an issue with the template, or send a PR. See CONTRIBUTING.md.

Star history

If it saves you from one bad meeting, a ⭐ helps the next person find it.

murder-board