seed-plan

The Business Planner turns an idea into the document behind a raise. It sets up the startup's shared config (name, one-liner, the raise, the founder), lays out the nine sections every investor expects, and, once Claude and the founder have written them, stitches in the numbers the other skills found: the market size, the competitors' weak points, the model's revenue, breakeven and runway. It works out the unit economics itself and checks the whole thing for holes: an empty section, a raise that is $1.5M in one file and $1.2M in another, a use of funds that does not add up, a market share bigger than the market. A plan is a way to think clearly, not a promise of funding: the pack does the work, investors decide.

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

python3 "${CLAUDE_SKILL_DIR}/plan.py" init --name "Leashly" --one-liner "A dog-walking app with flat prices and a backup walker on every walk" \
  --raise 1.5M --instrument "post-money SAFE" --founder "Your Name" --email you@leashly.example \
  --tagline "Dog walks without the surprise fees."
python3 "${CLAUDE_SKILL_DIR}/plan.py" build                 # seed/plan.md + the numbers in seed/plan.json
python3 "${CLAUDE_SKILL_DIR}/plan.py" check                 # the checks only: exit 0 clean, 1 something to fix

Other flags: --dir PATH (default ./seed), init --brand/--ink/--accent #RRGGBB (the deck's colours), init --address, --sector, --geography, --stage, init --force (start again), build --today YYYY-MM-DD, --json.

What it reads and writes

file	what
config.example.json (this folder)	the shape of seed/config.json, filled in for Leashly
seed/config.json	init writes it: the startup (name, one-liner, sector, stage, geography, raise, instrument), the founder (name, email, company address), sec_user_agent and the outreach limits (daily cap 20, follow-ups after 3 and 7 business days, 3 touches). Every seed skill reads it
seed/plan.json	init writes the skeleton; Claude writes each section's text, pitch (one line for the slide) and points (up to 3), plus tagline, team and brand. build adds numbers, flags, notes and the build date
seed/market.json	from /seed-market: TAM, SAM, SOM and their sourced inputs
seed/competitors.json	from /seed-scout: each app's rating, ratings, low reviews and weak point
seed/model.json	from /seed-model: the assumptions, the 36 months and the summary
seed/plan.md	build writes it: the plan with a key-numbers table, the checks and the nine sections with their tables

The unit economics

Computed from the model's assumptions, exactly, and shown in the plan:

number	how
revenue per order	price x take rate
contribution per order	revenue per order minus the costs of one order (card fees, per-order costs)
CAC payback	CAC / (contribution per order x orders per customer a month), in months
LTV	contribution per customer a month / monthly churn
LTV:CAC	LTV / CAC (most seed investors look for 3x or more)

The checks

• Flags (fix before anyone reads it): an empty section; a missing market.json, model.json or competitors.json; a raise that disagrees between config.json and model.json; a use of funds that does not add up to the raise; a SOM above the SAM or a SAM above the TAM; year 3 GMV above the SOM (the model sells more than the market you said you could win); a model.json that no longer matches its own assumptions; a contribution at or below $0.
• Notes (worth a look): a number in a section's text that is in none of the files (years and counts of 10 or less are skipped); LTV:CAC under 3x; CAC payback over 12 months; cash that runs out inside 36 months; no breakeven; no tagline, no team, an empty pitch line.

Steps

1. Run plan.py init with the startup's name, one-liner, the raise and the founder's own name and work email. The email is the one outreach goes from: never use anyone else's.
2. Run /seed-scout, /seed-market and /seed-model if their files are missing. The plan's numbers come from them, not from memory.
3. Write the nine sections into seed/plan.json with the founder, one at a time. Plain words, short paragraphs. Every number you write should exist in one of the files; if the founder gives a new number, add it to the right file first (the market inputs, the assumptions).
4. Give each section a pitch: one line, a slide's worth (the deck's lint allows 35 words a slide). Give the solution and go-to-market up to 3 points. Add the team (real people only) and a tagline.
5. Run plan.py build. Read back the key numbers and every flag and note.
6. Fix the flags (in the file the flag names), then build again until it says every check passed. Never change a number to silence a check without the founder's yes.
7. Hand off to /seed-deck to turn the plan into the 10-slide deck.

Example: the reel

Leashly (fictional): "A dog-walking app with flat prices and a backup walker on every walk", raising a $1.5M seed on a post-money SAFE to launch in 3 cities. plan.py build reads the reel's competitor numbers (Rover 4.9 stars and 515,771 ratings, 106 of 341 recent 1 and 2 star reviews mention fees; Wag! 4.8 stars and 67,447 ratings, 81 of 324 mention cancellations), the market (71M US homes with a dog, $158B US pet spend in 2025, a top-down TAM of $8.5B, SAM $228M, SOM $22.8M) and the model:

Leashly: business plan, 9 of 9 sections written.
  Per walk     price $25.00, take 15% = $3.75, contribution $2.23
  LTV:CAC      4.24x, CAC payback 3.37 months
Every check passed: no empty sections, and the numbers agree between files.
Wrote seed/plan.md and seed/plan.json.
Next: /seed-deck turns plan.json into the 10-slide deck.

The plan's key numbers: revenue $467,400 / $1,586,640 / $2,993,520 in years 1 to 3, breakeven in month 29 (May 2029), peak burn $82,801 a month, and the cash lasts all 36 months (low point $252,553). Year 3 GMV ($19.96M) sits under the $22.8M SOM, so that check passes.

Rules

• Never promise funding, in the plan or in the chat. Raising money depends on the business; investors decide.
• Projections come from the founder's own assumptions. Say so wherever the plan shows revenue or runway. Not investment advice.
• No invented numbers: every figure traces to a file, and every market input carries its source.
• The team slide and section list real people with their real roles.