Add ship-* Claude skills and plan

Introduces a skills/ directory with 8 Claude skills that reimagine ship
as a set of composable, human-driven deployment tools backed by Claude's
reasoning rather than a rigid CLI.

Skills:
- ship-setup: one-time VPS config, saves host to ~/.config/ship/config.json
- ship-status: derives live state from server, no local state file
- ship-env: read/write env vars with merge semantics, never overwrites
- ship-binary: deploy Go binaries with SQLite backup, correct restart behavior
- ship-caddy: manage per-app Caddyfile with validate-before-reload
- ship-service: systemd management and log inspection
- ship-static: rsync static sites with SPA routing support
- ship-deploy: orchestration runbook tying the others together

Also adds SKILLS_PLAN.md documenting the architecture and rationale.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

1 files changed, 126 insertions, 0 deletions

diff --git a/skills/ship-deploy/SKILL.md b/skills/ship-deploy/SKILL.md
new file mode 100644
index 0000000..60cc263
--- /dev/null
+++ b/skills/ship-deploy/SKILL.md
@@ -0,0 +1,126 @@
+---
+name: ship-deploy
+description: Deploy an app to a ship VPS. Orchestrates ship-binary or ship-static depending on what you're deploying. Use when you want to deploy or redeploy an app and want Claude to guide the process.
+argument-hint: "<app-name> [host-nickname]"
+---
+
+# ship-deploy
+
+Orchestration runbook for deploying apps to a ship VPS. Guides the full deployment
+process by calling the appropriate ship skills in the right order.
+
+## Prerequisites
+
+- `/ship-setup` must have been run at least once
+- `~/.config/ship/config.json` must exist and contain at least one host
+- The app must already be built locally (binary compiled or dist folder ready)
+
+## Read Config
+
+```bash
+python3 -c "
+import json, os
+cfg = json.load(open(os.path.expanduser('~/.config/ship/config.json')))
+nick = '<nickname-or-default>'
+h = cfg['hosts'].get(nick, cfg['hosts'][cfg['default']])
+print(h['host'])
+print(h['domain'])
+"
+```
+
+## Step 1 — Understand what we're deploying
+
+Ask the user (or infer from context):
+
+- **What is the app name?** (e.g. `foodtracker`)
+- **What type?** Binary or static site?
+- **Where is the artifact?** Path to binary or dist folder
+- **What host?** Default unless specified
+- **Any env vars needed?** Especially for first-time deploys
+- **Custom domain?** Or use `<app-name>.<base-domain>`
+
+If any of these are unclear, ask before proceeding.
+
+## Step 2 — Check if app already exists
+
+```bash
+ssh <host> "test -f /etc/ship/ports/<app-name> && echo exists || echo new"
+```
+
+Tell the user whether this is a fresh deploy or an update to an existing app.
+
+## Step 3 — Deploy
+
+### For a binary app → follow ship-binary
+
+Key steps in order:
+1. Backup any SQLite databases in `/var/lib/<app-name>/`
+2. Allocate or retrieve port
+3. Upload binary via scp
+4. Write/merge env file
+5. Write systemd unit
+6. `systemctl restart` (existing) or `systemctl enable --now` (new)
+7. Write Caddy config and reload
+
+### For a static site → follow ship-static
+
+Key steps in order:
+1. Validate `index.html` exists in dist folder
+2. Rsync dist folder to `/var/www/<app-name>/`
+3. Fix ownership to `www-data`
+4. Write Caddy config and reload
+
+## Step 4 — Verify
+
+After deploying, confirm the service came up:
+
+**Binary:**
+```bash
+ssh <host> "sudo systemctl is-active <app-name>"
+```
+
+If not active, immediately check logs:
+```bash
+ssh <host> "sudo journalctl -u <app-name> -n 30 --no-pager"
+```
+
+**Static:**
+```bash
+ssh <host> "curl -sI https://<domain> | head -5"
+```
+
+## Step 5 — Confirm to user
+
+Report:
+- App name and live URL
+- Type (binary / static)
+- New deploy or update
+- Port (binary only)
+- Any SQLite backups made
+- Any env vars that were set
+- Whether the service is running
+
+## Checklist (reference)
+
+Use this to make sure nothing is missed:
+
+- [ ] Config file read, host resolved
+- [ ] App type confirmed (binary / static)
+- [ ] Artifact path confirmed and exists locally
+- [ ] App name and domain confirmed
+- [ ] Existing app check done
+- [ ] SQLite backed up (binary, if db files exist)
+- [ ] Port allocated or retrieved
+- [ ] Artifact uploaded
+- [ ] Env file written with merge semantics
+- [ ] Systemd unit written and service started/restarted (binary)
+- [ ] Caddy config written and reloaded
+- [ ] Service confirmed running
+
+## Notes
+
+- Never skip the SQLite backup step for binary apps — always check even if you don't expect a db
+- Always merge env vars — never overwrite the whole env file
+- If anything fails mid-deploy, tell the user exactly where it failed and what state the server is in
+- Use `systemctl restart` for existing apps, `enable --now` for new ones — not the other way around
+- If the user says "deploy X" without more context, ask the minimum necessary questions before starting