flag-monitoring

Set up and manage a monitored progressive rollout (also called a "safe rollout") for a GrowthBook feature flag. A safe rollout is a standard rollout rule with a multi-step ramp schedule and monitoringConfig attached — the monitoring watches guardrail metrics at each step and can signal or automatically trigger a rollback if regressions are detected.

All API calls go through the bundled helper: ${CLAUDE_PLUGIN_ROOT}/scripts/gb-call. It needs GB_API_KEY set in env or written to ~/.config/growthbook/.env by /growthbook:gb-setup.

Required inputs for monitoring configuration

Before configuring, collect:

• Datasource ID — the datasource that tracks exposure and metric events
• Exposure query ID — which assignment query identifies users in the rollout
• Guardrail metric IDs — at least one metric that must not regress (e.g., error rate, crash rate)
• Signal metric IDs (optional) — leading-indicator metrics to watch, not hard gates
• SRM action — what to do on a Sample Ratio Mismatch: "hold" (recommended — pause for inspection) or "rollback" (aggressive) or "warn"
• Auto-rollback (autoUpdate in the monitored ramp payload, autoRollback in safe-rollout) — true means the system rolls back without human approval on guardrail failure; false holds for human review. Default to true unless the user has concerns about query cost or wants to control the cadence of monitoring snapshots manually

# Resolve datasource and exposure query IDs:
gb-call GET /api/v1/data-sources