UseCaseify GTM

Give your coding agent a GTM brain.

Your agent can build anything. It just doesn't know what to build next to grow.

This is not another GTM prompt pack — it is an experiment framework for coding agents (Claude Code, Codex, Cursor) with kill criteria locked before you start and a public registry of real experiments with real outcomes.

$ cd my-saas && claude
> /gtm-next-experiment

① inspects your actual repo — stack, analytics, signup flow
② asks you three questions
③ picks your ONE riskiest GTM assumption (ranked against alternatives)
④ designs ONE executable experiment and writes it into your repo:

   gtm/exp-001/BRIEF.md            ← for you
   gtm/exp-001/TASKS.md            ← for your agent — implements it right there
   gtm/exp-001/criteria.lock.json  ← continue / iterate / kill, frozen
   gtm/exp-001/reminder.ics        ← results day, on your calendar

…14 days later…

> /gtm-decision
⑤ collects the real numbers, judges against the LOCKED rules
   → DECISION.md: CONTINUE / ITERATE / KILL, with honest confidence

Runs on your agent and your tokens. No account required for local mode.

Install

From the root of your product's repository:

curl -fsSL https://raw.githubusercontent.com/llmjp/usecaseify-gtm/main/install.sh | bash

One command, three agents:

• Claude Code — skills land in .claude/skills/ → /gtm-next-experiment, /gtm-decision
• Codex / Cursor (and any AGENTS.md reader) — a marked routing block in AGENTS.md points to agent-neutral instructions in .agents/usecaseify-gtm/ → just say "run the next GTM experiment"

The script only writes those paths — read it first if you like. Other agents: bootstrap prompt.

Why this exists

Every founder using a coding agent has the same loop: ask an AI "how do I grow?", get fifty plausible ideas, implement none of them properly, learn nothing.

This framework enforces the discipline that actually produces learning:

1. One assumption at a time. The riskiest one, chosen against explicit criteria — business impact, uncertainty, cost of being wrong, testability, cost, time to evidence.
2. Criteria locked before the experiment starts. Continue / Iterate / Kill conditions are written down first, so the goalposts cannot move after you see the data.
3. Evidence honesty. Every claim is labeled — stated by you, observed in your repo, or AI inference. Confidence is qualitative and capped automatically when data is weak. Numbers that cannot be read from real data are reported as unknown, never estimated.
4. The human decides. The verdict is a recommendation. Overrides are recorded with a reason, next to the AI's recommendation — honestly.

The full method: docs/methodology.md · File format & JSON schema: docs/format.md · Real example (our own product): examples/xbackup-motivation-survey/

Two modes

	Local (no account)	Connected (free API key)
Generation	Your agent, your tokens, fully offline	Same — plus an optional hosted mode
Criteria lock	Local criteria.lock.json (self-hashed)	Server-notarized timestamp — provably locked before results existed
Your experiment page	—	Public page on the registry (preregistration, in public)
Case enrichment	—	Similar real experiments injected into generation (planned)

Get a free key at gtm.usecaseify.com — export USECASEIFY_API_KEY=... and the skills use it automatically.

The deal, stated plainly: free connected experiments are public — that's how the registry grows and why it's useful. Private experiments will be a paid tier. Reading the registry is free for everyone, forever. Fully-local mode asks nothing and reports nothing.

What lands in your repo

Everything is files, under version control. Your team can read, review, and diff every experiment — and your kill criteria live in git history where nobody can quietly rewrite them.

What this is not

• Not a marketing copy generator, an ad platform, or a growth-hack listicle.
• Not a guarantee. Experiments produce evidence; the decision stays yours.
• Not a data grab: local mode sends nothing anywhere. Connected mode sends what you publish — and rejects anything that looks like a credential.

Status

• Skills (local mode + hosted mode): working today
• Server-notarized locks & public registry pages: live — browse the registry
• usecaseify-gtm CLI (validate / render): planned
• Case enrichment from the registry: planned

Built in public by LLM JAPAN, LLC — the team behind UseCaseify. Questions: support@usecaseify.com

License

MIT