shark-board

The first meeting. Before anyone builds a website or signs a lease, five board members tear the idea apart, each from a different angle, and vote. The board is there to find what kills the idea, not to cheer for it.

Read ${CLAUDE_SKILL_DIR}/../receipts/rules.md first.

What the user typed after /shark-board: $ARGUMENTS. If that is blank, take the idea from the conversation.

The rule about the five names

The lenses are summaries of published frameworks, nothing more. The board members are not those people, never speak as them, never quote them, never claim they would endorse anything. Write "the Offers lens", never "Alex Hormozi says". If the user asks what one of them really thinks, say you can only apply their public frameworks.

Step 0: the idea file

Everything in the Shark Tank Proof pack reads founder/idea.md in the user's project. If it does not exist, write it now from the conversation:

# <business name, or a working name>

- What it is: one sentence.
- Who it is for: the target customer, as specific as the user can make it.
- What it sells, at what price: the core product and its price.
- Where and how: city or online, how customers find it, how it is delivered.
- Budget and constraints: money, time, skills, anything already decided.
- Stage: Stage One (no sales yet, or under about $1 million a year) unless the
  user says otherwise, with yearly sales if there are any.
- What they sell next: the second product for a customer who bought the first,
  or "none yet".

If the name, the customer, the product or the price is missing, ask for those (at most three questions in one message), then write the file. Do not invent a budget, a market size or a competitor.

Step 1: brief the board

Write founder/board/brief.md: the idea file plus anything else the user said that a stranger would need. Sub-agents cannot see this conversation, so the brief must stand on its own. Do not add your own opinion of the idea to it.

Step 2: convene

Launch five sub-agents in ONE message (Agent tool, subagent_type: general-purpose, one per member), each told: Read founder/board/brief.md and shark-board's lenses.md (path below), apply ONLY the <lens> lens, and write your memo to founder/board/<lens>.md. Your job is to find what would kill this business. Do not be agreeable: most new businesses fail, and a polite FUND is worse than a clear PASS. The founder's excitement is not evidence. The Evidence member also reads founder/evidence/claims.md if it exists. The lens file is lenses.md in this skill's folder (${CLAUDE_SKILL_DIR}/lenses.md; if that path is not expanded, use the base directory Claude Code printed for this skill).

Each memo, in this shape:

# <Lens> lens
Score: <1-10> | Vote: FUND / FUND IF / PASS
## What is strong
## What would kill it (the top 3, most dangerous first)
## The cheapest real test for each (what, with whom, by when, what result kills it)
## The first sale: where the first customers are, what they buy first, at what price
## What is real so far, and what is only believed
## The questions I would ask the founder
## If FUND IF: the conditions (each one a real-world test, not "do more research")

Step 3: the chair's summary

Read the five memos (they are short) and write founder/board.md:

0. The hard truth: the single most likely reason this business dies, in one or two plain sentences. This comes first, before the vote.
1. The vote, e.g. "2 fund if, 2 pass", and the average score.
2. The risks more than one member raised: these go first.
3. Every condition, merged, as a checklist. Mark which ones only real buyers can tick off (most of them), and which the CFO or the competitor scout can. The simulated panel can raise doubts about a condition; it can never tick one off.
4. The kill list: the three tests to run first in the real world, with the result that would mean stop. Put the riskiest assumption first, tested on its own, and prefer tests that can run within two weeks (Ready, Fire, Aim, rule 11 in rules.md).
5. The stage and the first sale: which stage the business is in, and the board's best answer to the four selling questions (where, what first, what price, what words). Mark each answer as a guess until a real buyer acts.
6. One paragraph: the strongest version of this business the board can see. It may differ from what the user pitched; say so plainly.

Then log the vote in the ledger as SIMULATED (it is model opinion):

python3 "${CLAUDE_SKILL_DIR}/../receipts/ledger.py" add --kind other --status SIMULATED --use PRIVATE --by shark-board --claim "Board vote: <vote>, average <score>/10"

Keep the members' disagreements. A board that agrees on everything was not doing its job.

Rules

• Score the business, not the founder.
• No made-up facts. If a lens needs a number nobody has yet (market size, cost of acquiring a customer), name it as an open question and say which skill answers it (/ground-truth, /truth-serum, /burn-test).
• A PASS is a result, not a failure. Say it plainly.
• If the founder argues with a vote, the board changes it only for new evidence, and the chair names that evidence. Arguing harder is not evidence.
• No member treats the panel, another memo or the founder's belief as proof.

Output

founder/idea.md, founder/board/ (brief and five memos), founder/board.md. Then say the hard truth and the vote, one line each, and suggest the next step: the first kill-list test with real people (/receipts has the call guide), /ground-truth to see who is already there, or /truth-serum to find objections with simulated buyers before the real calls.

Where the pack goes next

evidence -> board -> competitors -> consumer -> pricing -> offer -> cfo -> marketing -> brand -> ops -> launch -> plan -> pitch -> funding

/receipts also runs any time a real person does something: a call, a deposit, a pilot, a no.

Every skill can also run on its own; each reads whatever is already in founder/.