seed-coach

The Pitch Coach gets you ready for the questions that sink first calls. It reads the numbers the other seed skills wrote and turns them into the questions a sharp investor will ask about them, with your own figures next to each one so you can answer in a sentence. Then it grades what you say against the things investors listen for, and lays out the 20 minutes of a booked call so you spend them on the story, not on small talk.

${CLAUDE_SKILL_DIR} below is this skill's folder (where this SKILL.md lives).

python3 "${CLAUDE_SKILL_DIR}/coach.py" drill --n 10 --customers owners --provider walker
python3 "${CLAUDE_SKILL_DIR}/coach.py" drill --investor maya-chen --category "the ask"
python3 "${CLAUDE_SKILL_DIR}/coach.py" drill --category "unit economics" --seed 3 --write
python3 "${CLAUDE_SKILL_DIR}/coach.py" grade --answer "15% flat, no owner fee. ..." --question ue-take
python3 "${CLAUDE_SKILL_DIR}/coach.py" grade --file answer.txt
python3 "${CLAUDE_SKILL_DIR}/coach.py" agenda --investor maya-chen --write

Other flags: --dir (default ./seed), --json, --minutes 10..60 for the agenda, --seed N to shuffle within each category the same way every time.

What it reads and writes

How it works

• Drill. A tailored question is asked only when the files hold every number it needs; otherwise it is skipped, never filled with a guess. Tailored questions come first, one per category per round, then the generic ones, so ten questions cover the ground. --customers and --provider set the words ("owners", "walker pay").
• Grade, five checks of 20 points: 15 to 150 words (under 60 seconds at 150 words a minute), a number, no hedging words ("I think", "maybe", "probably"), at least two specifics (numbers, names) and no buzzwords ("disrupt", "world-class"), and the answer first (a first sentence of 25 words or fewer). Exit 0 only when all five pass.
• Agenda: seven blocks, 2, 3, 4, 4, 4, 2 and 1 minutes for 20 (scaled exactly for other lengths): hello and why them, the problem (the review evidence), the product, market and model, their questions (the three most likely), the ask, the next step.

Steps

1. Run coach.py drill --n 10 with the user's words for customers and the supply side. Ask the questions one at a time, as the investor would.
2. Let the user answer in writing or out loud (transcribe it), then run coach.py grade --answer "..." --question ID. Read the unchecked lines back and ask for a tighter version. Two or three rounds per question.
3. When a numbers question stumps them, show the "Your numbers" line and the file it came from. If a number looks wrong, fix it in that skill (/seed-model, /seed-market), not in the answer.
4. When /seed-closer books a call, run coach.py agenda --investor ID --write and drill --investor ID, and walk through both the day before.
5. After the call, hand back to /seed-closer to log it and send the follow-up, and to /seed-dataroom if they ask for documents.

Example: the reel

Leashly (made up) has a call with Maya Chen (fictional) of Northlight Ventures on Thursday, November 5, 2026 at 10:00 AM ET. coach.py drill --customers owners --provider walker opens with:

 1. [market] Your SOM is 2.5% of a $960M SAM in 3 years. What gets you the first 1,000 customers?
 2. [competition] Rover has 515,771 ratings. Why will owners switch?
 3. [unit economics] What does your 15% take rate leave after walker pay?
    Your numbers: Per walk: $25 price, you keep $3.75, $2.23 after costs.

The answer "15% flat, no owner fee. We keep $3.75 of every $25 walk and $2.23 after card fees and the backup walker. Rover keeps 20% of sitter bookings and adds an owner fee." grades 100/100: 32 words, about 13 seconds. coach.py agenda --investor maya-chen opens with her personal line, puts Rover's 106 of 341 and Wag!'s 81 of 324 in minutes 2 to 5, and ends on the $1.5M post-money SAFE and a date for the next meeting.

What it never does

It never invents a number for an answer, never tells the user to overstate traction, and never promises that a good call means a check. Investors decide.