Flag Graph

flag-graph

Trace the dependency relationships around a GrowthBook feature flag. Use this skill before making structural changes to a flag (renaming, archiving, deleting) to understand the blast radius, or when building a mental model of how flags depend on each other.

Read-only — this skill never writes.

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:setup.

Workflow

1. Fetch the flag

gb-call GET /api/v2/features/

Capture: prerequisites (feature-level), rules (check for rule-level prerequisites, experiment-ref entries, safe-rollout entries), holdout.

2. What does this flag depend on? (forward dependencies)

From the flag's prerequisites array, for each prerequisite:

gb-call GET /api/v2/features/

Show: prerequisite flag ID, its current state (enabled envs, default value), and the condition the current flag is checking against it. Recurse one level if the prerequisite also has prerequisites — surface the full chain, noting where it ends.

Also check rule-level prerequisites in the rules array — each rule can have its own prerequisites field. Surface these as rule-scoped dependencies.

3. What depends on this flag? (reverse lookup)

GrowthBook has no reverse-prerequisite API endpoint. A full reverse lookup requires scanning all flags:

gb-call GET /api/v2/feature-keys

Then paginate through all flags looking for any that list the target flag in their prerequisites:

gb-call GET '/api/v2/features?limit=100'
gb-call GET '/api/v2/features?limit=100&offset=100'   # continue until exhausted

For each flag returned, check prerequisites[*].id and rules[*].prerequisites[*].id against the target flag ID.

Warn the user: this is an O(n) scan across all flags. On large orgs with hundreds of flags, it may take several paginated calls.

4. Which experiments link to this flag?

Check for experiment-ref rules in the flag's rules array (field experimentId on rules with type: "experiment-ref"). For each:

gb-call GET /api/v1/experiments/

Surface: experiment name, status (running/stopped/draft), and whether the flag's id is the experiment's trackingKey.

Also check if any experiments list this flag in linkedFeatures:

gb-call GET '/api/v1/experiments?trackingKey='

This catches experiments wired by convention (experiment-launch sets trackingKey === flag-id). Complement with the experiment-ref rule scan above for experiments wired manually.

5. Holdout associations

If the flag has a holdout field set, note the holdout ID and warn: "This flag participates in holdout ``. Deleting or significantly changing this flag could affect holdout analysis."

6. Present the dependency report

Dependency graph for ``:

DEPENDS ON (forward):
  → flag-Y (prerequisite, condition: value === true)
      → flag-Z (flag-Y's prerequisite, condition: value === "v2")

DEPENDED ON BY (reverse, scanned  flags):
  ← flag-A (feature-level prerequisite)
  ← flag-B (rule-level prerequisite on rule "Beta testers")
  [limitation: reverse lookup scanned all flags; may miss any created after this scan]

EXPERIMENTS:
  exp_abc123 "Checkout experiment" — status: running, trackingKey matches

HOLDOUTS:
  holdout_xyz — flag participates; remove cautiously

SAFE TO DELETE?: 

Guardrails

• Reverse lookup is a full scan — rate-limit aware. The 60 rpm rate limit applies. If the org has >600 flags, the scan takes 10+ API calls. Surface the count before starting and offer to proceed.
• Reverse lookup is point-in-time. Flags created after this scan won't appear. Note the scan timestamp in the report.
• No API for reverse prerequisite lookup. There is no GET /features?dependsOn= endpoint. The scan is the only reliable approach.
• Experiment trackingKey scan catches the common case. experiment-launch sets trackingKey === flag-id by convention. Manual wiring (different trackingKey, linked via linkedFeatures) may not be caught by the trackingKey query — the rule scan in step 4 is the defensive check.
• Read-only. This skill never writes. For changing prerequisites, use flag-prerequisites. For deletion, use flag-cleanup.
• Circular dependency detection. If during the forward-dependency traversal you encounter a flag that points back to the starting flag, surface it as a circular dependency warning — it means the flag can never fully evaluate.

Endpoints used

• GET /api/v2/features/:id — fetch flag and its dependencies
• GET /api/v2/feature-keys — full flag ID list for reverse lookup
• GET /api/v2/features (paginated, limit/offset) — full flag scan for reverse lookup
• GET /api/v1/experiments/:id — fetch linked experiment details
• GET /api/v1/experiments?trackingKey= — find experiments linked by convention

Handoffs

• flag-prerequisites — to add, remove, or modify feature-level prerequisites
• flag-cleanup — to archive or delete the flag after confirming the blast radius
• experiment-stop — if a linked experiment is running and needs to be stopped before flag removal
• flag-search — to find all flags matching criteria (broader than single-flag dependency tracing)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: growthbook
• Source: growthbook/skills
• License: MIT

Install and usage instructions live in the source repository linked above.