Part 9: Production Deployment and Sanity Skill

Deployment should be treated as an engineering event, not just a button click.

Skill for This Post: `deploy-and-sanity`

Create .ai/skills/deploy-and-sanity/SKILL.md:

---
name: deploy-and-sanity
description: Execute production deployment with preflight checks, rollout controls, and post-deploy sanity verification.
---
 
# Deploy and Sanity
 
## Preflight
- Verify branch, commit hash, and release notes
- Confirm QA and UAT signoff
- Validate migration and rollback plans
- Confirm monitoring dashboards and alert channels
 
## Rollout
- Use staged rollout if available
- Track errors, latency, saturation, and key business metrics
- Pause or rollback if thresholds are breached
 
## Post-Deploy Sanity
- Validate top 5 critical user journeys
- Check logs and traces for new error signatures
- Confirm background jobs/events are healthy
- Publish release status summary
 
## Required Output
1) preflight status
2) rollout status
3) sanity-check results
4) final status: success | partial | rollback
5) follow-up actions

Prompt Pattern

Run deploy-and-sanity for release vX.Y.Z.
Show preflight, rollout, and sanity results.
If any critical metric regresses, recommend rollback.

Why This Skill Matters

Reduces risky ad-hoc release behavior
Gives teams a consistent production checklist
Makes incident response faster when issues appear

Cross-Agent Install

ln -sfn "$PWD/.ai/skills/deploy-and-sanity" ~/.claude/skills/deploy-and-sanity
ln -sfn "$PWD/.ai/skills/deploy-and-sanity" ~/.codex/skills/deploy-and-sanity

Wrap-Up

You now have a full skill-driven delivery system:

mode selection (including when not to default to gstack)
workflow orchestration
PRD intake
task breakdown
implementation
code review
QA
UAT
production deployment and sanity

Start small: pick one skill, run it for one sprint, then expand.