Flag Ramp

flag-ramp

Create and manage multi-step ramp schedules for a GrowthBook feature flag rule. A ramp schedule progressively increases (or decreases) a rule's traffic coverage over time, with defined hold intervals between steps. Steps can be manual (operator advances them) or time-gated (auto-advance after an interval).

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.

How ramp schedules work

A ramp schedule is attached to a specific rule on a feature flag. When published, the ramp begins running immediately — there is no automatic delay unless you explicitly set a startDate (uncommon; most teams prefer to trigger the ramp manually after publish via the GrowthBook UI).

At each step, the schedule applies actions — patches to the rule, typically changing coverage. After a step:

• If the step has an interval (seconds), it auto-advances after that duration.
• If interval is null, it holds until a team member manually advances it in the UI.

When all steps complete, endActions are applied (typically setting coverage to 1.0). If the ramp is rolled back at any point, startActions are applied (restoring the pre-ramp state).

Ramp schedules are staged on a draft revision as rampActions and executed atomically at publish time.

Workflow

Path A — Create a ramp schedule on an existing rule

1. Identify the rule:

gb-call GET /api/v2/features/

Show the rules list. Get the ID of the force or rollout rule the user wants to ramp. Note current coverage.

2. Design the ramp steps with the user:

Confirm:

• Starting coverage (e.g., 0.05 = 5%)
• Step progression (e.g., 5% → 25% → 50% → 100%)
• Hold time at each step: a number in seconds (e.g., 86400 = 1 day) to auto-advance, or null to hold until manually advanced

Hold-for-approval steps: add holdConditions: { "requiresApproval": true } to any step that needs human sign-off before the ramp advances. Approval is always the final gate — if the step also has an interval, the timer must elapse first, then the step waits for an explicit approval call. A pure manual gate (no time component) is { "interval": null, "holdConditions": { "requiresApproval": true } }. Mixed examples:

• Soak for a day, then require approval: { "interval": 86400, "holdConditions": { "requiresApproval": true } }
• Require approval only, no time hold: { "interval": null, "holdConditions": { "requiresApproval": true } }

See Path D for how to submit approvals via API once the interval has cleared.

startActions should match the rule's current coverage (captured in step 1) — this is the state the ramp restores to on rollback.

3. Build the ramp schedule payload:

{
  "startActions": [
    { "targetType": "feature-rule", "targetId": "", "patch": { "coverage": 0 } }
  ],
  "steps": [
    {
      "interval": 86400,
      "actions": [{ "targetType": "feature-rule", "targetId": "", "patch": { "coverage": 0.1 } }]
    },
    {
      "interval": 86400,
      "holdConditions": { "requiresApproval": true },
      "actions": [{ "targetType": "feature-rule", "targetId": "", "patch": { "coverage": 0.5 } }]
    }
  ],
  "endActions": [
    { "targetType": "feature-rule", "targetId": "", "patch": { "coverage": 1.0 } }
  ]
}

startActions = the state applied when the ramp begins and restored on rollback. Set this to the rule's current coverage before the ramp starts — typically 0 for a new rule. endActions = final state after all steps complete (typically 100% coverage).

To add guardrail metric monitoring to the ramp, add a monitoringConfig block and "monitored": true on each step — see flag-monitoring.

Omit startDate and cutoffDate unless the user explicitly requests them — see Guardrails.

4. Attach the ramp schedule to the rule via draft:

echo '' \
  | gb-call PUT /api/v2/features//revisions/new/rules//ramp-schedule -

Capture the returned version. The ramp schedule is staged as a rampAction on the draft — it becomes live when the draft is published.

5. Hand off to flag-publish.

Path B — Create a new rule with a ramp schedule in one step

When creating the rule and ramp together, fetch available hashAttribute candidates first (required for rollout rules):

gb-call GET '/api/v1/attributes?projectId='

Validate the rule's value against the flag's valueType (captured from the flag fetch). Then post:

echo '{
  "rule": {
    "type": "rollout",
    "value": "",
    "coverage": 0.05,
    "hashAttribute": "",
    "enabled": true,
    "allEnvironments": false,
    "environments": [""]
  },
  "rampSchedule": {
    "steps": [
      {
        "interval": 86400,
        "actions": [{ "targetType": "feature-rule", "targetId": "new", "patch": { "coverage": 0.05 } }]
      },
      {
        "interval": 86400,
        "actions": [{ "targetType": "feature-rule", "targetId": "new", "patch": { "coverage": 0.25 } }]
      },
      {
        "interval": 86400,
        "actions": [{ "targetType": "feature-rule", "targetId": "new", "patch": { "coverage": 1.0 } }]
      }
    ],
    "startActions": [{ "targetType": "feature-rule", "targetId": "new", "patch": { "coverage": 0.0 } }],
    "endActions": [{ "targetType": "feature-rule", "targetId": "new", "patch": { "coverage": 1.0 } }]
  }
}' | gb-call POST /api/v2/features//revisions/new/rules -

Use "targetId": "new" as a placeholder in the ramp actions — the server replaces it with the actual rule ID on creation.

Path C — Remove a ramp schedule from a rule

gb-call DELETE /api/v2/features//revisions/new/rules//ramp-schedule

This stages a detach ramp action on the draft. The live schedule is removed when the draft publishes. The rule's coverage is left at whatever the schedule last set it to.

Path D — Manage a live ramp (post-publish)

Before taking any action, suggest the user open the feature page for a full picture of ramp progress — the UI surfaces schedule status, step timeline, and metric health that the API only returns as raw values:

# macOS:
open /features/
# Linux:
xdg-open /features/

After that, for API-based management: get the ramp schedule ID from the flag's rules (rampScheduleId field on the rule), or look it up:

gb-call GET '/api/v1/ramp-schedules?featureId='
# or more precisely:
gb-call GET '/api/v1/ramp-schedules?ruleId='

Check status:

gb-call GET /api/v1/ramp-schedules//status

Returns: current step index, decision ("advance" / "hold" / "rollback" / "waiting"), health signals, and whether the step is awaiting approval.

Start the ramp (if it hasn't started automatically):

gb-call POST /api/v1/ramp-schedules//actions/start

Approve a hold-for-approval step (after the interval has elapsed):

gb-call POST /api/v1/ramp-schedules//actions/approve-step

Returns 400 if the interval hasn't cleared yet or other holds are still active — poll /status first and only approve when the step is awaiting approval.

Advance to next step (override all holds — use for CI pipelines or hard overrides):

echo '{}' | gb-call POST /api/v1/ramp-schedules//actions/advance -
# Force past an unsatisfied approval gate (requires canBypassApprovalChecks permission):
echo '{"force":true}' | gb-call POST /api/v1/ramp-schedules//actions/advance -

Pause / Resume:

gb-call POST /api/v1/ramp-schedules//actions/pause
gb-call POST /api/v1/ramp-schedules//actions/resume

Roll back (restores startActions state, marks as rolled-back):

echo '{"reason":""}' | gb-call POST /api/v1/ramp-schedules//actions/rollback -

Complete immediately (skip remaining steps, apply endActions):

gb-call POST /api/v1/ramp-schedules//actions/complete

Restart after rollback:

gb-call POST /api/v1/ramp-schedules//actions/restart
# then start again:
gb-call POST /api/v1/ramp-schedules//actions/start

Emergency kill-switch (faster than rollback): disable the flag environment via flag-toggle — kills all rules instantly without needing the ramp schedule ID.

Guardrails

• Draft version threading. If a version number is already in context from a previous write skill in this session, use it explicitly instead of new. Fall back to new when starting fresh.
• Ramp schedules apply only to force and rollout rules. They don't work on experiment-ref or safe-rollout rules.
• startActions must match the rule's pre-ramp coverage. Capture it in step 1 and use it as the rollback state. Don't default to 0 unless the rule was at 0 before the ramp.
• Steps with null interval hold until manually advanced in the UI. Useful for human-gated checkpoints. Steps with an interval auto-advance.
• startDate is optional — the ramp starts immediately on publish if omitted. Only ask for it if the user specifically wants a delayed start; most teams prefer to trigger manually after verifying the publish succeeded.
• cutoffDate is a niche safety net. Only mention it if the user asks — it's a deadline that auto-rolls the ramp back if reached. Don't include it by default.
• Coverage patches on steps must be within 0–1. The server validates this at publish time.
• Check the target environment is enabled. If the flag is disabled in the target env, the ramp will do nothing — warn and route to flag-toggle first.
• For monitored ramps (guardrail metrics, auto-rollback signals), use flag-monitoring.
• Ramp actions are staged on the draft. Nothing changes until published. If the draft is discarded, the ramp is never created.
• One ramp schedule per rule. Adding a new one replaces any existing schedule.

Endpoints used

Draft (pre-publish):

• GET /api/v2/features/:id — fetch flag and current rules
• PUT /api/v2/features/:id/revisions/new/rules/:ruleId/ramp-schedule — stage ramp schedule on a draft
• DELETE /api/v2/features/:id/revisions/new/rules/:ruleId/ramp-schedule — stage ramp detach
• POST /api/v2/features/:id/revisions/new/rules — create rule with inline rampSchedule

Live ramp management:

• GET /api/v1/ramp-schedules — list (featureId, ruleId, status filters)
• GET /api/v1/ramp-schedules/:id/status — real-time health and decision
• POST /api/v1/ramp-schedules/:id/actions/start
• POST /api/v1/ramp-schedules/:id/actions/pause
• POST /api/v1/ramp-schedules/:id/actions/resume
• POST /api/v1/ramp-schedules/:id/actions/advance (body: optional { force: true })
• POST /api/v1/ramp-schedules/:id/actions/approve-step
• POST /api/v1/ramp-schedules/:id/actions/rollback (body: { reason: string })
• POST /api/v1/ramp-schedules/:id/actions/restart
• POST /api/v1/ramp-schedules/:id/actions/complete

Handoffs

• flag-monitoring — to add guardrail metrics and automated monitoring signals to the ramp
• flag-targeting — to set up the rule's targeting conditions before attaching a ramp
• flag-toggle — for an emergency kill-switch if the ramp needs to be stopped immediately
• flag-publish — to publish the draft and activate the ramp

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.