Ship

Ship Check

Run a go/no-go checklist before deploying. Catches build failures, leaked secrets, debug artifacts, missing env vars, broken tests, and other landmines that slip through when you're shipping fast and solo.

Sample output

A successful run produces something like:

## Ship Check — marketing-site
Deploy target: Vercel
Branch: main (clean, up to date with origin)

PASS Git state — clean working tree
PASS Build — npm run build passed (12.4s)
PASS Tests — 47 passed, 0 failed
PASS Type check — clean
FAIL Secrets — found "sk_live_" in src/lib/stripe.ts:14 (should use env var)
FAIL Env vars — NEXT_PUBLIC_GA_ID set locally but not in Vercel dashboard
WARN Debug — 2 console.log calls in src/app/checkout/page.tsx
WARN TODO — 1 unresolved TODO in src/checkout.ts:28 (added in this branch)

### Blockers (do not deploy)
1. Hardcoded Stripe secret key — move to env var STRIPE_SECRET_KEY
2. Missing NEXT_PUBLIC_GA_ID in Vercel — analytics will silently fail in prod

### Warnings (deploy with awareness)
1. console.log statements left in src/app/checkout/page.tsx
2. Unresolved TODO in this branch's diff

Verdict: NO-GO. Fix blockers, re-run /ship.

Arguments

The user may specify:

• No args: auto-detect the project from the current working directory
• A project name or path (e.g., myapp, ~/code/marketing-site)
• --skip-tests: skip the test suite (useful when tests are slow and you've already run them)
• --skip-build: skip the build step (useful when you just built successfully)

Parse the argument string. Examples:

• /ship -- full check on current directory
• /ship myapp -- full check on myapp
• /ship --skip-tests -- skip test suite
• /ship myapp --skip-build -- skip build for myapp

Project Aliases

| Alias | Path | |-------|------| | myapp | (add your projects here) |

Process

Step 0: Detect the project

1. Resolve the project path (alias table or cwd)
2. Read package.json, pyproject.toml, Cargo.toml, or equivalent to identify the stack
3. Read CLAUDE.md if it exists for project-specific deploy instructions or conventions
4. Detect the deploy target: check for vercel.json, railway.json, fly.toml, .github/workflows/deploy*, Procfile, netlify.toml, or eas.json

Step 1: Git state

Run these checks:

• Uncommitted changes: git status. Flag any unstaged or staged-but-uncommitted changes. These won't be in the deploy.
• Unpushed commits: git log @{u}..HEAD --oneline (if tracking a remote). Flag commits that exist locally but haven't been pushed. If the deploy triggers from a push, these are the changes that will ship.
• Branch check: What branch are you on? If it's not main/master and the project deploys from main, warn. If it's a feature branch, note it.
• Merge conflicts: Search for >>>>>> in tracked files. These should never ship.

Step 2: Build

Skip if --skip-build is passed.

Detect and run the build command:

| Signal | Build command | |--------|--------------| | package.json with build script | npm run build or yarn build or pnpm build | | pyproject.toml or setup.py | Check for build/compile steps in project docs | | Cargo.toml | cargo build --release | | Makefile with build target | make build | | Static HTML (no build step) | Skip, note "no build step" | | Expo/React Native (eas.json) | Skip build (EAS builds are expensive). Just verify the config. |

Pass: Build exits 0 with no errors. Fail: Build exits non-zero. Show the error output. Stop here — nothing else matters if the build is broken. Warning: Build succeeds but has warnings. Show them. Common warnings that matter: unused exports (tree-shaking issues), large bundle size warnings, deprecation warnings from dependencies.

Step 3: Tests

Skip if --skip-tests is passed.

Detect and run the test command:

| Signal | Test command | |--------|-------------| | package.json with test script | npm test | | pytest.ini or pyproject.toml [tool.pytest] | pytest | | Cargo.toml | cargo test | | .rspec | bundle exec rspec | | Go files with _test.go | go test ./... |

Pass: All tests pass. Fail: Any test fails. Show the failure. This is a hard stop. No tests: Note "no test suite found." This is a warning, not a blocker — but flag it. Production apps should have tests.

Step 4: Secrets and sensitive data

Search the tracked files (not node_modules, not .git, not build output) for:

Hard stops (these must not ship):

• Hardcoded API keys: patterns like sk_live_, sk_test_, pk_live_, AKIA, ghp_, gho_, glpat-, xoxb-, xoxp-
• .env files that are tracked (check git ls-files for .env, .env.local, .env.production)
• Private keys: -----BEGIN (RSA |EC |DSA )?PRIVATE KEY-----
• Database connection strings with credentials: postgres://user:pass@, mongodb+srv://user:pass@
• Hardcoded passwords: password = ", PASSWORD = ", secret = " (with a non-empty string value, not a variable reference)

Warnings (review these):

• console.log / console.debug / console.warn / print() / debugger statements in non-test source files. Exclude:
• Test files (*.test.*, *.spec.*, __tests__/)
• Config files (*.config.*)
• Server-side logging that's intentional (look for structured loggers like winston, pino, logging.getLogger)
• Lines that are clearly error handling (console.error in a catch block)
• TODO or FIXME comments in files that changed in the current branch (compared to main). Old TODOs in untouched files are not your problem right now.
• Commented-out code blocks in files that changed. Shipping commented-out code is sloppy.

Step 5: Environment variables

Check that the project's expected env vars are accounted for:

1. Find env var references: Search source code for process.env., os.environ, os.getenv, env::var, import.meta.env. patterns. Collect the variable names.
2. Find env var definitions: Check .env.example, .env.local.example, vercel.json (env section), railway.json, fly.toml, docker-compose.yml, and any documented env var lists in README or CLAUDE.md.
3. Compare: Flag any env var that's referenced in code but has no definition in any example/config file. This is a var that a fresh deploy might be missing.
4. Check .env.example freshness: If .env.example exists, check whether all vars referenced in code appear in it. Flag missing ones.

This check does NOT read actual .env files (those contain secrets). It only compares what code expects vs. what's documented.

Step 6: Dependency check

Quick sanity check:

• Lockfile exists: Does the lockfile match the package manager? (package-lock.json for npm, yarn.lock for yarn, pnpm-lock.yaml for pnpm, Pipfile.lock for pipenv, poetry.lock for poetry). Missing lockfile = non-deterministic installs.
• Lockfile in sync: Run npm ls --all 2>&1 | grep "missing" or equivalent. If the lockfile is out of sync with the manifest, the deploy might install different versions than what you tested.
• No file: or link: dependencies: These work locally but break in CI/deploy.

Don't run a full vulnerability audit here — that's what /vuln is for. Just check the basics.

Step 7: Stack-specific checks

Next.js

• Check next.config.* for output: 'export' if deploying to static hosting
• Search for use client in files that use server-only features (or vice versa)
• Check for generateStaticParams on dynamic routes if using static export

Astro

• Check astro.config.* for output mode matching deploy target
• Verify integrations are installed (e.g., @astrojs/sitemap in config but not in dependencies)

Python (Flask/Django)

• Check DEBUG = True or debug=True is not hardcoded in production config
• Check SECRET_KEY is not hardcoded
• Check ALLOWED_HOSTS or equivalent is configured for production domain

Expo / React Native

• Check app.json / app.config.js for correct bundle identifier and version
• Check eas.json profiles match intended build type
• Do NOT trigger an EAS build — those cost money. Just verify the config.

Static HTML

• Check for broken relative links (href="./ or src="./ pointing to files that don't exist)
• Check that index.html exists

Step 8: Report

Present a clear go/no-go:

## Ship Check -- [project name]
**Branch:** main (3 commits ahead of origin/main)
**Stack:** Next.js 16 / Vercel

### BLOCKERS (must fix before deploy)
- [ ] Build failed: [error summary]
- [ ] Merge conflict markers in src/components/Header.tsx
- [ ] Hardcoded Stripe key in src/lib/stripe.ts:14

### WARNINGS (review before deploy)
- [ ] 3 console.log statements in changed files
- [ ] TODO in src/checkout.ts:28 (added in this branch)
- [ ] ENV var RESEND_API_KEY referenced in code but not in .env.example

### CLEAN
- [x] Git state: 0 uncommitted changes
- [x] Tests: 14 passed, 0 failed
- [x] Dependencies: lockfile in sync
- [x] Secrets: no leaked keys
- [x] Stack checks: all clear

### VERDICT: [GO / NO-GO / CAUTION]

Verdicts:

• GO -- zero blockers, zero warnings. Ship it.
• CAUTION -- zero blockers, some warnings. Probably fine, but review the warnings.
• NO-GO -- one or more blockers. Fix them first.

Rules

• Blockers are hard stops. Failed build, failed tests, leaked secrets, merge conflicts. These are non-negotiable.
• Warnings are judgment calls. A console.log in a debug utility might be intentional. A TODO in a non-critical path can wait. Report them, let the user decide.
• Don't fix anything. This skill reports. It doesn't modify code. If something needs fixing, say what and where — the user (or another skill) handles the fix.
• Be fast. Skip slow operations when the user asks (--skip-tests, --skip-build). A ship check that takes 5 minutes won't get used.
• Respect project conventions. If CLAUDE.md says "we don't write tests for this project," don't flag the lack of tests as a warning. If the project has a custom deploy process, note it.
• Don't duplicate other skills. This is not /vuln (no full vulnerability scan), not /zombie (no dead code audit), not /seo (no SEO checks), not /autopsy (no test quality review). This is strictly "is this safe to deploy right now."
• Scope to what's shipping. When checking for TODOs, console.logs, and commented-out code, focus on files that changed in the current branch vs. main. Legacy junk in untouched files is not a deploy blocker.

Skill Run Logging

After completing this skill, append a log entry to ~/.claude/skill-log.md under the Run Log table. New entries at the top (most recent first).

Format: | YYYY-MM-DD | /skill-name | target (details) | one-line result |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Surfrrosa
• Source: Surfrrosa/claude-skills
• License: MIT
• Homepage: https://shainapauley.com/skills

Install and usage instructions live in the source repository linked above.