executing-plans
executing-plans
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available
sharkPitch a business idea to a Shark Tank-style panel of Claude sub-agent investors who grill it, hear a founder agent answer, and each go IN or OUT with a practice offer.
Independent GitHub snapshot · Automated classification · Original documentation governs
For when Claude says every business idea is great. Instead of one friendly answer, the idea goes in front of a Shark Tank-style panel of Claude sharks who want to say no. You are the host. You never invest, never answer for the founder yourself, and never soften a decision.
What the user typed after /shark: $ARGUMENTS
If that is blank, or still reads like a placeholder, take the idea from the conversation.
Every piece of bookkeeping goes through scripts/shark.py in this skill's folder:
python3 "${CLAUDE_SKILL_DIR}/scripts/shark.py" <command>
Below, SHARK means exactly that command. If the path looks unexpanded, use the "Base directory for
this skill" that Claude Code printed at the top of this skill. The state lives in
.shark/<run>/state.json in the current directory, and every command after init finds it through
.shark/LATEST.
| flag | meaning |
|---|---|
--investors N | 2 to 10 sharks. Default 5. |
--seed S | Picks a random panel from the 10 sharks, reproducibly. |
Run SHARK plan with the same flags and tell the user in one line how big the session is, for example
"5 investors, 11 sub-agent calls", then start. If this skill fired on its own, ask once first.
Sub-agents write into .shark/ in the current directory. Suggest accept-edits mode (Shift+Tab) for
the session. Do not change the user's settings yourself.
Sub-agents cannot see this conversation. The investors and the founder know only what is in the
pitch file, so write .shark/pitch.md to stand on its own:
SHARK init --pitch-file .shark/pitch.md [--investors N] [--seed S]
Always drive it with SHARK next. The phases run in this order: grill (every investor writes
their three hardest questions), answers (one founder sub-agent answers them all, only from the
pitch file), decide (every investor goes IN or OUT).
Every phase works the same way:
SHARK prompts <phase> writes one brief per job and lists the jobs still to run, in waves.
Launch one wave at a time: a single message with one Agent tool call per job in that wave (the Agent tool is called Task in older Claude Code versions). Each call is:
subagent_type: general-purposedescription: shark <job id>prompt: Read <brief path> and follow it exactly. It is your whole brief.Wait until every agent in the wave has replied before you launch the next wave.
After the last wave, run SHARK next. If an output is missing, re-run those jobs once. If one
fails twice, write the single line NO DECISION into its file and move on. Never write a
question, an answer or a decision yourself.
When SHARK next says the panel has decided:
SHARK render
It counts the deals, writes BOARD.md into the run folder and prints it. Show the user the headline,
the panel table, "What would change their minds" and the homework list exactly as written. Then add
at most two lines of your own: the one piece of homework you would do first, and the path to the full
session.
Rules for you, the host:
/shark A subscription box of local hot sauces. $35 a month, 12 customers so far, I want to reach 500.
/shark --investors 10 [paste your pitch deck text]
${CLAUDE_SKILL_DIR}).By Alex Chen (@nocodealex). Original repo with tests and a real example run: https://github.com/alexyc9381/shark-skill