Deploy to Server

Deploy a project to production safely: check first, deploy, verify after.

Config

Server credentials live in ~/.jarvis/servers.md (never in this skill, never in git):

| Name | Host | User | Auth | Projects |
|------|------|------|------|----------|
| prod | 1.2.3.4 | deploy | ssh key | my-app |

If the file doesn't exist, ask the user for server details once and create it.

Before Deploy — ALWAYS

1. Blackout check — if the user defined no-deploy hours (e.g. business hours for a client-facing app), refuse and say when the next window opens
2. Tests — run the project's test suite; a red suite blocks the deploy
3. Uncommitted changes — git status must be clean; deploy from a pushed commit, not from the working tree

Deploy

Docker projects

ssh $USER@$HOST "cd ~/$PROJECT && git pull && docker compose build && docker compose up -d"

Bare projects (systemd / supervisor)

ssh $USER@$HOST "cd ~/$PROJECT && git pull && sudo systemctl restart $PROJECT"

Prefer pull-based deploys (server pulls from git) over pushing files with scp — the server state stays reproducible.

After Deploy — ALWAYS

1. Tail logs for errors: docker logs --tail 30 $PROJECT or journalctl -u $PROJECT -n 30
2. Hit the health endpoint: curl -sf https://myapp.com/health — non-200 means roll back
3. Report to the user: what was deployed, commit hash, health status

Rules

• Never kill -9 — always graceful restart
• Never edit code directly on the server
• If the health check fails, roll back first, debug second