experiment-brainstorm

Propose new experiment ideas grounded in the team's past stopped experiments. Read history first; propose based on what actually moved metrics, where guardrails failed, and which tags or projects under-explored.

All API calls go through the bundled helper: ${CLAUDE_PLUGIN_ROOT}/scripts/gb-call. It needs GB_API_KEY — set in your shell, or written to ~/.config/growthbook/.env by /growthbook:gb-setup. If unset or invalid, gb-call's error message points back at /growthbook:gb-setup.

Workflow

Pull the experiment list, most recent first.
```
gb-call GET '/api/v1/experiments?limit=50&status=stopped&sortBy=dateCreated&sortOrder=desc'
```
Returns up to 50 experiments per page (limit caps at 100). Keep sortBy=dateCreated&sortOrder=desc — the API's default order is oldest-first, and on an org with more than one page of history an unsorted pull grounds every proposal in ancient experiments.

If the user scoped the brainstorm ("ideas for checkout", "what should the growth team test next"), narrow the pull with filters instead of discarding results after the fact: &tag=checkout, &projectId=prj_abc123, &owner=<email>, &result=won,lost, &metricId=met_abc123, or &implementationType=feature. tag, owner, result, metricId, and implementationType take comma-separated values (ORed within a param); projectId takes a single project id. Separate params AND together.

Consider adding &bandits=false when the user wants product-test patterns: bandit results carry per-arm probabilities rather than a winner/loser verdict, so they distort the win-rate tally in step 4.