growthbook — Agent Skill

GrowthBook Agent Skills

GrowthBook is an open source Feature Flagging and Experimentation platform.

Agent skills for GrowthBook — feature flagging and experimentation playbooks for Claude Code, Cursor, and other agent tools that follow the Agent Skills standard.

The skills call the GrowthBook REST API directly through a small bundled helper. No MCP server required.

What's included

Four skills, one per domain. Each one is a router: it loads a short index, then reads only the workflow that matches what you asked for. The 27 workflows live as reference files inside their skill and stay out of context until they're needed.

feature-flags

The full flag lifecycle. Flag changes go through a draft revision before going live, so the draft → review → publish workflows sit alongside the ones that make the change.

| Workflow | What it does | | --- | --- | | flag-create | Create a new feature flag — collision check, value type, environments, defaultValue. Ships disabled everywhere. | | flag-search | Search, list, and audit flags by project, tag, owner, environment state, or staleness. Read-only. | | flag-graph | Trace a flag's dependency graph: prerequisites, dependents, linked experiments and holdouts. Read-only. | | flag-toggle | Enable or disable a flag in a specific environment (the kill switch). Review-gated. | | flag-targeting | Add, edit, or remove force / rollout rules — with conditions, saved groups, and rule-level prerequisites. Full operator reference for MongoDB-style conditions. | | flag-rules | List rules, delete a rule, reorder, or route to the right rule workflow. | | flag-experiment | Add an experiment-ref rule to a flag to run an A/B test through it. | | flag-schedule | Time-gate a rule: set a start and/or end datetime for automatic activation. | | flag-ramp | Multi-step ramp schedule: progressively increase coverage over time with per-step intervals or manual approval gates. Includes live ramp management (advance, pause, rollback, approve-step). | | flag-monitoring | Monitored progressive rollout ("safe rollout"): ramp schedule with guardrail metric monitoring and optional auto-rollback. | | flag-prerequisites | Gate an entire flag on another boolean flag being on. | | flag-default-value | Change the fallback value served when no rules match. | | flag-metadata | Update a flag's description, owner, project, tags, custom fields, or JSON schema. | | flag-revisions | List and inspect open drafts, check who owns them, see approval status, create or discard drafts. The "what's in flight?" workflow. | | flag-review | Request an approval review on a draft, or submit one (approve / request-changes / comment). | | flag-publish | Publish a draft live, resolve merge conflicts (rebase), discard, or revert to a prior revision. | | flag-cleanup | Archive or delete a stale flag, walking through code-site inlining first. Two-step safety gate (archive → verify → delete). |

experiments

| Workflow | What it does | | --- | --- | | experiment-brainstorm | Propose new experiment ideas grounded in your team's past stopped-experiment history. | | experiment-design | Walk through hypothesis, variations, primary metric, guardrails, and sample size to produce a launchable spec. Reads only. | | experiment-launch | End-to-end launch: create the experiment, prep or reuse the feature flag, wire the experiment-ref rule, and call /start. Handles approval and pre-launch checklist failure paths. | | experiment-analyze | Trigger a fresh snapshot, poll until ready, then interpret results (SRM check, lifts, CIs, guardrails). | | experiment-stop | Stop a running experiment, optionally declaring a winner and enabling a temporary rollout. Full post-stop flag disposition guidance. | | learnings | Search, read, and record Learnings — durable conclusions drawn across multiple experiments. Check before designing a test; record after one generalizes. |

analytics

Turn the metrics and fact tables you already use for experimentation into ad-hoc charts and saved dashboards with GrowthBook's Product Analytics Explorer.

| Workflow | What it does | | --- | --- | | metric-search | Search, list, and audit fact metrics and fact tables — definitions, columns, and what's chartable. Read-only. | | metric-create | Create a fact metric, and the fact table underneath it when one doesn't exist yet. Covers all seven metric types; analysis settings inherit the org defaults. | | analytics-explore | Build and run a chart: a metric over time, a fact-table aggregation, or a raw warehouse table. Returns the numbers plus a deep link to the rendered chart. | | dashboard-create | Build a dashboard from a goal or a set of metrics. Chart blocks carry a config and the create call runs every one of them, so it's one write for the whole page. | | dashboard-edit | Change a dashboard that exists: add or remove a chart, swap a metric, change the timeframe, rename or re-scope it. Reads the dashboard, then writes the full block list back. |