poly-gents Feature flagging
Purpose
Guide safe progressive delivery using feature flags: cohort sizing, success/rollback criteria, kill switches, and flag lifecycle hygiene.
When to Use
- User asks about feature flags, gradual rollout, kill switch, or flag cleanup.
- Planning canary/percentage rollouts with observability guardrails.
- Separating release flags from entitlements/permissions.
When NOT to Use
- A/B experiment design and statistics → ab-experimentation.
- Infrastructure feature stores you do not have context for—gather vendor/docs first.
- Code implementation without flag context → implement-features.
Expected Outcome
- Rollout plan with cohort, metrics, rollback/kill steps, and flag lifecycle (owner, expiry).
- Clear distinction between release toggles and permission flags.
Inputs to Gather
- Flag name/key, default state, and targeting rules.
- Success metrics and alert thresholds.
- Owner, expiry/removal ticket, and environments.
- On-call kill-switch procedure.
Workflow
- Classify flag type (release vs permission vs experiment).
- Define default-off/on rationale and smallest first cohort.
- Set success metrics and rollback triggers before expanding.
- Document kill-switch steps for on-call.
- Plan removal ticket to avoid long-lived debt.
- Pair with monitor-slos / observability for guardrails during rollout.
Domain guidance
- Prefer small cohorts first; define success and rollback criteria before expanding.
- Separate release flags from permission or entitlements.
- Every flag needs: owner, expiry or removal ticket, default-off or default-on rationale.
- Document kill-switch steps for on-call.
Avoid long-lived flags without cleanup plans.
Tool Availability Rules
| Access | Behavior |
|---|---|
| Full tool access | Update flag config via integrated tools when approved. |
| Read-only | Produce rollout plan and ticket text. |
| No flag platform | Architectural guidance only; no fabricated flag state. |
Related tool sets
githubjira
Review / Decision / Execution Criteria
- Rollback path tested or documented before wide rollout.
- No permission checks hidden behind release flags without review.
- Expiry or removal ticket required for new flags.
Output Format
- Flag inventory (in scope).
- Rollout phases and cohorts.
- Metrics and rollback/kill triggers.
- Lifecycle (owner, expiry).
- Risks and open questions.
Quality Bar
- On-call can execute kill switch from your doc alone.
Safety and Boundaries
- Confirm production flag changes affecting all users.
- Do not fabricate current flag targeting state.
Escalation / Dispatch Rules
- Experiment analysis → ab-experimentation.
- Incident during rollout → lead-incidents.
References
skills/old_skills.json(feature-flagging).skills/skill.instruction.md,skills/meta.instructions.md