deploy

name	deploy
description	Deploy phase. Push to the deployment target, verify it's live, write README and DELIVERY.md.
disable-model-invocation	true

Deploy the verified software to its target.

Steps

1. Read scaffolding/scope.md for deployment target

2. Read scaffolding/design.md for deployment config details

3. Pre-flight check — before attempting deploy, verify access for the target in scope.md:

   Generic checks (all targets):

   • Deploy CLI/tool is installed (which <tool>)
   • Authenticated to the deploy target (logged in, token available)
   • Deploy config file exists (e.g., docker-compose.yml, Dockerfile, GitHub Actions workflow, platform config)
   • Required env vars / secrets are set (check existence without printing values)
   • Build succeeds before attempting deploy

   Platform-specific examples:

   • GitHub Pages: git remote -v shows valid remote with push access
   • Docker / VPS: SSH access to host, Docker installed, docker-compose.yml exists, .env on host
   • Container registry: Registry credentials available in env, Dockerfile builds
   • Binary release: Build completes, artifact exists at expected path

   If any pre-flight check fails: STOP. Do not attempt deploy. Report what's missing.

4. Identify the rollback command before deploying. Write it down:

   If this deploy breaks, rollback command: [specific command]
   

   Examples:

   • GitHub Pages: git push origin main (re-triggers build from last good state)
   • Docker / VPS: docker compose up -d --no-deps <service> with previous image tag pinned in .env or compose file
   • Container registry: docker tag <prev> <latest> && docker push or redeploy previous image
   • Binary release: Previous release remains on GitHub — users can grab it

   The rollback command goes into DELIVERY.md under "Incident Response."

5. Deploy to the specified target:

   • GitHub Pages: Build, push to gh-pages branch or configure Actions
   • Docker / VPS: Build image, docker compose up -d on host (via SSH or CI)
   • Container registry: Build image, push, deploy
   • Cron/script: Set up the schedule, verify it runs
   • Local/manual: Document exact run commands

6. Verify it's accessible and working

7. Write README.md in the project root with:

   • What this is (one paragraph)
   • How to set up locally
   • How to deploy
   • How to run tests

8. Write DELIVERY.md in the project root. This is the client-facing handoff document. All projects get the same structure — depth scales naturally with the project's complexity.

DELIVERY.md format

# Delivery — [Project Name]

## What Was Built

[What it does, who it's for — 1-3 sentences]

## How to Use It

[Run command, URL, or access method]

## Acceptance Criteria — Verified

- [x] [Criterion 1] — Evidence: [how it was verified]
- [x] [Criterion 2] — Evidence: [...]

## Deferred Items

[Items from scope.md Deferred section. If none: "None — full scope delivered."]

## Known Limitations

[What it doesn't do, edge cases, performance bounds]

## Operational Notes

[How to monitor, restart, check logs, common issues. For simple tools: "Run [command]. No persistent state."]

## Architecture Overview

[Brief summary of how the system is structured. Point to scaffolding/design.md for full details.]

## Data & Migrations

[Current schema version, migration instructions, backup strategy. If stateless: "Stateless — no data persistence."]

## Security Posture

[Auth method, data handling, input validation, compliance notes. At minimum: "No secrets in code, HTTPS, parameterized queries."]

## Incident Response

[What to do when things break. Point to RUNBOOK.md if it exists. For simple tools: "Check logs, restart process, report issue."]

## Support Terms

[Bug-fix period, contact method, what's included vs. out of scope]

## Next Steps / Roadmap

[Recommended future work from deferred items, known limitations, and client input. If final: "No further work planned."]

## Version History

[v1 — date — summary of what was delivered]
[v2 — date — summary of iteration changes (if applicable)]

For simpler projects, sections will naturally be shorter (one-liners). The agent should not pad sections with filler — if a section is "N/A" or a single sentence, that's fine. The structure stays consistent so every delivery is navigable the same way.

Post-Deploy Gate

• Deployed to specified target
• Accessible (can reach it, run it, or verify the cron fires)
• README.md exists with setup + run + deploy instructions
• DELIVERY.md exists with at minimum: what was built, how to use it, known limitations
• If stateful: data persistence verified (can create, read back)

If any gate condition fails, fix it and recheck. Up to 3 retries.

Log the result to scaffolding/log.md with URL/access method as evidence.

Git checkpoint:

git add -A && git commit -m "chore(deploy): deploy to [target]" -m "[URL or access method]\nGate: post-deploy PASS (attempt N)."

STOP here and report to the user:

✓ FULL PIPELINE COMPLETE.
[URL or access method]
[What was deployed and where]
[Summary of scaffolding/log.md — all phases, all gate results]

DELIVERY.md written with handoff details.
Scaffolding and input docs persist as project provenance.
To iterate: add feedback to docs/input/ and run /iterate.