-
Notifications
You must be signed in to change notification settings - Fork 1
Workflow Release Readiness
Pre-release checklist + rollout playbook. Runs as Stage 11 of Workflow PRD to Production, after validation has passed.
For the canonical version, see workflows/release-readiness.md.
- Validation report has
Status: PASS(see Workflow Testing and Validation). - Staging environment runs the release candidate stably for the duration specified in the PRD.
- No active P0 / P1 incidents.
[ ] Validation report attached to the PRD
[ ] Staging environment stable for >= duration specified in PRD
[ ] Performance budgets met in staging load tests
[ ] Rollback procedure drilled and confirmed working
[ ] Feature flag wired and tested (on / off)
[ ] Observability dashboards updated
[ ] Alerts configured for the new feature's failure modes
[ ] Documentation merged (PRD Section 12)
[ ] Changelog / release notes drafted
[ ] On-call team briefed on the new surface
[ ] Customer comms drafted (if user-visible change)
[ ] Compliance / privacy sign-off (if required by feature scope)
[ ] No active P0/P1 incidents
[ ] Deployment owner identified
Every box must be checked. Skipping a box and adding "we'll handle it after deploy" is the failure mode this checklist exists to prevent.
PRD Section 19 defines the strategy. Common shapes:
- 10% of traffic for 24 hours; monitor.
- 50% of traffic for 24 hours; monitor.
- 100% of traffic; monitor for 7 days.
At each stage:
- Monitor: error rate, latency p95/p99, success metrics from PRD Section 3.3.
- Trigger rollback if any metric breaches the threshold in PRD Section 19.1.
- Deploy to one region or one cluster first.
- Monitor.
- Promote to remaining regions.
- Deploy everywhere at once.
- Only appropriate when blast radius is minimal and rollback is fast.
For most features, gradual rollout is the default. Big-bang requires explicit ADR.
- Traffic shift confirmed in load balancer / feature flag.
- Error rate within baseline + 0.1%.
- p95 latency within budget.
- Alerts not firing.
- Customer impact: zero P0/P1 reports related to the feature.
If all green for 24 hours, advance. Otherwise rollback.
- Same metrics as 11a with higher floor.
- Cache hit rates stabilizing as expected.
- Database load within projected envelope.
If all green for 24 hours, advance. Otherwise rollback.
- Full traffic.
- Monitor for 7 days.
- Compare success metrics (PRD Section 3.3) against targets.
If success metrics within target after 7 days, declare release complete. Otherwise: investigate gap, fix, re-validate.
Per PRD Section 19.1:
- p95 latency >
<threshold>sustained for 5 minutes. - Error rate >
<threshold>sustained for 5 minutes. - Customer-reported correctness regression with reproduction.
Triggers are mechanical. The on-call engineer doesn't deliberate — if a trigger fires, rollback is the response.
Per PRD Section 19.2:
- Toggle feature flag to off (or revert deploy if no flag).
- Verify error rates drop to baseline within 1 minute.
- Open an incident ticket.
- Investigate root cause.
- Fix and re-validate before re-attempting rollout.
-
Update PRD status to
Shipped. Move fromprds/active/toprds/archive/. -
Update spec triplet status to
Complete. -
Capture lessons learned in a brief retrospective. The
prd-authoragent can draft a starting outline. - Identify follow-ups. Anything explicitly out of scope (PRD Section 4.2) that's now worth doing → new PRD.
| Milestone | Notify |
|---|---|
| 10% rollout begins | Engineering team, product owner |
| 50% rollout begins | Engineering team |
| 100% rollout begins | Engineering team, product owner, support |
| 100% stable for 7 days | Engineering team, product owner, support, customers (if user-visible) |
| Rollback triggered | All of the above + on-call escalation |
Comms templates are project-specific. Drafting is part of release readiness; ad-hoc comms during incidents are noisy.
- Skipping the rollback drill. Untested rollback is no rollback.
- Big-bang for non-trivial changes. Saves a day; costs a week if it goes wrong.
- Skipping the 7-day soak at 100%. Some failure modes only surface at steady-state full traffic.
- Closing the PRD before metrics are validated. Premature closure leaves the PRD claim ("time-to-find < 2s") unverified.
- No on-call briefing. On-call gets paged at 3am for a feature they've never seen.
The deployment / SRE role (see agents/archetypes/devops-agent.md).
- Workflow PRD to Production — Stage 11 in context
- Workflow Testing and Validation — Stage 10 (the gate before)
Repository · Issues · Roadmap · Changelog · Security · License (Apache 2.0) — © Enovatr Labs
- Philosophy
- Spec-Driven Development
- Agentic Coding Model
- Automation Decision Framework
- Multi-Agent Orchestration
- Two-Tier Docs Pattern
- Multi-Vendor Context Files
- Documentation Structure
- Agent Memory
- Artifact Taxonomy
- PRDs
- Specs
- Agents
- Skills
- Commands
- Hooks
- Prompts
- Rules
- Frontmatter Contracts
- Sanitization