growthbook/skills

feature-flags

Read and modify GrowthBook feature flags — create, search, toggle, target, schedule, ramp, monitor, publish drafts, and archive.

Zobacz źródło
Oryginalny dokument Skill

Treść z repozytorium z zachowaniem nagłówków, przykładów, kodu, tabel, linków i obrazów.

feature-flags

Domain router for GrowthBook feature flags. Every flag workflow lives in a reference file under references/. Read this router, pick the workflow that matches the user's intent, then read that one file and follow it.

Flags use the v2 API (/api/v2/features). Environments, projects, and saved groups are v1.

All API calls go through the bundled helper. Under the Claude Code plugin install, it lives at ${CLAUDE_PLUGIN_ROOT}/scripts/gb-call (the plugin root). Under npx skills install, it lives at scripts/gb-call relative to this skill's directory. Resolve that path once and substitute it whenever a reference example says gb-call; do not assume gb-call is on PATH. It reads GB_API_KEY from the environment first, then falls back to ~/.config/growthbook/.env (written by gb-setup); environment variables take precedence.

When a workflow needs to present a GrowthBook UI link, a shell-capable runtime should call gb-call app-origin once per conversation, retain the returned trusted origin across workflow and domain handoffs, and prepend it to the root-relative UI paths in the reference file. If the origin is already in context, reuse it; do not call the command once per link. Embedded or MCP adapters that already know their trusted app origin may resolve the same paths directly. Never derive an app origin from GB_API_URL; if app-origin refuses because self-hosted configuration is incomplete, route to gb-setup.

Pick a workflow

Read exactly one of these, based on what the user is doing. If two look plausible, read the more specific one.

Read thisWhen the user wants to
references/flag-create.mdCreate a new flag
references/flag-search.mdFind, list, or audit flags — by project, tag, owner, environment state, or staleness (read-only)
references/flag-graph.mdTrace what a flag depends on, what depends on it, and its linked experiments or holdouts (read-only)
references/flag-toggle.mdEnable or disable a flag in an environment — the kill switch
references/flag-targeting.mdAdd, edit, or remove force / rollout rules, conditions, saved groups, rule-level prerequisites
references/flag-rules.mdList, reorder, or delete rules, or route to the right rule-type workflow
references/flag-experiment.mdAdd an experiment-ref rule so an A/B test runs through the flag
references/flag-schedule.mdTime-gate a rule with a start and/or end datetime
references/flag-ramp.mdBuild or drive a multi-step ramp schedule (progressive coverage over time)
references/flag-monitoring.mdSet up a monitored / safe rollout — a ramp plus guardrail metrics and optional auto-rollback
references/flag-prerequisites.mdGate the whole flag on another boolean flag
references/flag-default-value.mdChange the value served when no rules match
references/flag-metadata.mdChange description, owner, project, tags, custom fields, or JSON schema
references/flag-revisions.mdSee what drafts are open, who owns them, start or discard a draft
references/flag-review.mdRequest a review, or approve / request changes / comment on a draft
references/flag-publish.mdPublish a draft, resolve a merge conflict, revert, or discard
references/flag-cleanup.mdArchive or delete a stale flag, inlining its value at code sites first

Ambiguity worth resolving before you read: "turn off the flag" is flag-toggle if they mean the whole environment, flag-rules if they mean one rule. "Roll this out" is flag-targeting for a one-shot percentage, flag-ramp for a schedule, flag-monitoring if guardrail metrics should gate it.

Shared conventions

These hold across every flag workflow. The reference files assume them.

  • Everything writes through a draft revision. Rules, metadata, default value, and toggles all create or update a draft, which then has to be published. flag-publish owns the publish step; the write workflows hand off to it.
  • Thread the draft version through a chain. When several write workflows run in sequence in one session, carry the version returned by the previous step instead of re-sending new. new auto-creates or reuses a draft, which can silently pick up a teammate's concurrent draft mid-chain. Only fall back to new (or /revisions/latest?mine=true) on a fresh start.
  • POSTing `rules` replaces the entire array. The v2 rules array is top-level and scoped by environments or allEnvironments. GET the current rules first for any partial edit.
  • `defaultValue` is always a string, whatever the flag's value type.
  • Show users the flag key (the id), never an internal Mongo id. Link to the UI path /features/<key>.
  • UI links are root-relative paths. Never infer the GrowthBook UI origin from GB_API_URL; API and UI origins need not share a hostname. A runtime with a trusted app origin may resolve the path; otherwise surface the root-relative path without guessing.
  • Print deep links, don't open them. This skill has no browser grant on purpose — surface the path and let the user click when its runtime can resolve it.

Failure modes

Verified against the back-end handlers; don't collapse these into one another.

  • 400 on the publish endpoint means approval is required (This revision requires approval before publishing). The policy gate is working as intended — hand to references/flag-review.md.
  • 403 on a path that publishes as a side effect means the same thing, from a different code path: the environment toggle and POST /v2/features/<id> with {archived} both go through createAndPublishRevision, which raises a permission error rather than a bad-request error. Read the body — if it doesn't mention review, the token lacks permission outright and the problem is GB_API_KEY.
  • 409 on publish means the draft's base is stale — the live flag moved. Do not auto-rebase. Halt, show the conflict, let the user resolve it.
  • 422 on publish means a publish gate is blocking. The body lists which gates a plain ignoreWarnings retry would clear and which need a permission instead.

Read-only vs. write

flag-search and flag-graph are read-only and must stay that way — they never POST, PUT, or DELETE. Every other workflow writes, and each one states its own confirmation requirements. flag-cleanup's archive-then-verify-then-delete gate is a product-safety pause, not a mutation confirmation: don't skip it even when the user sounds certain.

Budget

GrowthBook rate-limits at 60 rpm. flag-search and flag-graph fan out across paginated lists — cap loops explicitly and mention the call budget when it's non-obvious.

Handoffs

  • The experiments skill — designing, launching, analyzing, or stopping an A/B test. A flag's experiment-ref rule is this skill's business; the experiment itself is not.
  • The analytics skill — charting product data or browsing metrics and fact tables.
  • gb-setup — when gb-call reports a missing or invalid GB_API_KEY.
z tego samego repozytorium

Więcej Skills

Wszystkie Skills
growthbook
Społeczność

analytics

Chart GrowthBook product data, build Analytics dashboards, manage the metric catalog, or query the warehouse directly — run Product Analytics explorations, save charts together on a dashboard, search metrics and fact tables, create fact metrics and their fact tables, or fall back to ad-hoc SQL. Use for "show me signups by country", "chart daily active users", "how many orders last week", "build me a dashboard", "put these metrics on a dashboard", "add a chart to this dashboard", "set up reporting for X", "find our revenue metric", "what fact tables exist", "create a metric", "define a metric on the orders table", "what tables contain user data", "run a SQL query", or any "show me / chart / plot / how many" question about product data. For an A/B test's results or choosing experiment metrics, use experiments. For feature flags, use feature-flags. For first-time API key configuration, use gb-setup.

instalacje
1
GitHub Stars
21
Aktualizacja
20 wrz
growthbook
Społeczność

experiments

Design, launch, analyze, and stop GrowthBook A/B tests, and search or record durable Learnings across experiments. Use for experiments, split tests, variations, hypotheses, sample size, guardrail metrics, SRM, chance to win, lift, declaring a winner, "what have we learned", "have we tested this before", or "record this learning". For bandits, identify them and direct the user to GrowthBook UI. For feature flags and rollouts, use feature-flags. For product analytics or metric catalog work, use analytics. For API key configuration, use gb-setup.

instalacje
1
GitHub Stars
21
Aktualizacja
20 wrz
growthbook
Społeczność

gb-setup

Configure GrowthBook API credentials and the trusted app origin so the other skills can run and link to the UI. Use when the user says "set up growthbook", "configure my api key", "growthbook isn't working", "where do I put my key", or when another skill emits an error pointing here ("GBAPIKEY is not set", "authentication failed", "GBAPPURL is required"). Writes ~/.config/growthbook/.env with chmod 600 and validates against the live API. For listing flags or running experiments, the domain skills handle that.

instalacje
1
GitHub Stars
21
Aktualizacja
20 wrz