ship-it

name	ship-it
description	Run the full CI-mirror locally, bump SW cache version, commit, and push. Use when you're ready to deploy a change to Geriatrics. Replaces the "edit → test → bump SW → commit → wait 60s for CI" loop.
disable-model-invocation	true

/ship-it

Mirrors what .github/workflows/ci.yml and integrity-guard.yml check, but locally. Fails fast, refuses to push on failure.

What it runs

1. npm test — full Vitest suite (all tests must pass, 0 failures)
2. Node syntax check on shlav-a-mega.html, sw.js, scripts/**/*.cjs, scripts/**/*.js
3. JSON schema validation on all data/*.json
4. Duplicate question ID detection
5. GRS-reference detection in data/notes.json
6. innerHTML audit on shlav-a-mega.html
7. SW cache version bump (auto-increments patch: v9.48 → v9.49)
8. git add the changed files
9. Prompts for commit message
10. git push origin main

Usage

/ship-it
/ship-it "feat: add 12 new fall-risk MCQs"     # skip commit-message prompt
/ship-it --dry-run                              # run checks, don't commit or push
/ship-it --no-bump                              # skip auto SW version bump

How I (Claude) should execute this

When the user invokes /ship-it, run:

bash .claude/skills/ship-it/ship.sh "$ARGS"

The script is non-interactive when given a commit message, interactive otherwise. Surface its output verbatim to the user — don't summarize. If it exits non-zero, DO NOT retry or "fix it up" — the failures it surfaces are real CI gates. Report what failed and ask the user how to proceed.

Design

• Fail loud. If any gate fails, stop. Don't try to be clever.
• No auto-fixes. Fixing a failing test or adding a missing citation isn't this skill's job — that's regular Claude editing.
• SW bump is opt-out, not opt-in. It's the single most-missed step; defaulting to auto-bump catches it.
• Push is the last step. So a test failure never leaves orphan commits unpushed vs. pushed.