ship-check
Run before anything you built with AI leaves your laptop. Scans a project folder (and its git history) for passwords, API keys and tokens, including the ones Claude Code saves into .claude/settings.local.json when you click "don't ask again". Then walks through the four checks a script cannot do. Use when the user says ship check, is it safe to publish, check this before I push, make this repo public, or is about to share an app, a repo or a skill.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/aobeidzinski-cloud--ship-check)HTML
<a href="/listings/aobeidzinski-cloud--ship-check"><img src="/api/listings/aobeidzinski-cloud--ship-check/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
1. Download this repo (Code, then Download ZIP) and unzip it. 2. Move the `ship-check` folder into your Claude Code skills folder: `~/.claude/skills/` 3. Open Claude Code in any project and say: `run a ship check on this folder`
No Claude Code? The script runs on its own:
bash scripts/ship-check.sh /path/to/your/project
It needs bash, grep, awk and find, which every Mac and Linux machine has. git is optional.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md · Manifest · 67 lines
- ---
- name: ship-check
- description: Run before anything you built with AI leaves your laptop. Scans a project folder (and its git history) for passwords, API keys and tokens, including the ones Claude Code saves into .claude/settings.local.json when you click "don't ask again". Then walks through the four checks a script cannot do. Use when the user says ship check, is it safe to publish, check this before I push, make this repo public, or is about to share an app, a repo or a skill.
- ---
- # Ship check
- A plain-language safety pass for people who build with AI and are not security engineers. It answers one question: if I publish this folder right now, what goes with it that I did not mean to share?
- ## When to use it
- - Before making a GitHub repo public or pushing to one that already is.
- - Before sharing a Claude Code skill, a script or a zip of a project.
- - Before publishing an app built with an AI builder.
- - After reading a news story about leaked keys and wondering about your own.
- ## Step 1: run the scan
- ```bash
- bash scripts/ship-check.sh /path/to/the/folder
- ```
- No folder means the current one. It needs bash, grep, awk and find. git is optional. It prints blockers and warnings and never prints a secret value: you see the file, the line and the first four characters.
- What it checks:
- 1. **AI agent settings files.** `.claude/settings.local.json` and `.claude/settings.json` hold every command you approved with "don't ask again", in full. If a command had a password or token typed into it, the file has it too. The scan counts those commands and tells you whether git has ever seen the file.
- 2. **Secrets in the files you are about to publish.** Common key shapes plus anything that looks like `password = ...` or `token: ...`.
- 3. **Secrets in git history.** Deleting a key from a file does not delete it from history, and a public repo publishes its whole history.
- 4. **Files that should never ship.** `.env`, `.pem`, `.key`, credentials files, and whether `.gitignore` covers them.
- 5. **Things that say more than you meant.** Paths from your computer (they show your username) and email addresses.
- Add your own terms, such as a client name, at run time. Do not save them in the script:
- ```bash
- EXTRA_RE='client-name|internal-project' bash scripts/ship-check.sh .
- ```
- Exit code 0 means no blockers, 1 means blockers, 2 means the folder was not found.
- ## Step 2: read the result with the user
- Go through every blocker and warning in plain words. For each one, say what it is, whether it looks real or like a placeholder, and what to do:
- - **A real credential, anywhere.** Rotate it first (make a new key at the provider, turn the old one off). Clean up second. A key that was public for a minute should be treated as stolen.
- - **A secret in history.** Do not try to fix history in place unless the user knows how. The safe route is a fresh folder with only the current clean files, a new repo, and the old one kept private.
- - **A settings file that git tracks.** `git rm --cached .claude/settings.local.json`, add it to `.gitignore`, and delete the rules that hold secrets.
- - **A placeholder that looks like a key.** Reword it to plain words such as `PASTE_YOUR_KEY_HERE`. Key-shaped placeholders get uploads blocked by GitHub's push protection.
- Never print, repeat or summarize a secret value. Never rotate, delete or commit anything without the user saying yes.
- ## Step 3: the four checks a script cannot do
- Ask the user each one and wait for the answer.
- 1. **Open it logged out.** If this is a live app, open it in a private window with no login. Click around. Whatever loads for a stranger, the whole internet can read. If the app has a database, ask the AI builder directly: "Which tables can a visitor read without signing in?"
- 2. **Check every share setting.** "Anyone with the link" means public. That includes the app, the docs it links to, dashboards and file folders.
- 3. **List what the agent can write to.** For every connector, token or MCP server the project uses: can it only read, or can it also create, edit and delete? Start on read-only. Add write access one tool at a time, when a task needs it.
- 4. **Approve patterns, not passwords.** When Claude Code asks to run a command with a key typed into it, do not choose "don't ask again". Put the key in an environment variable and approve the command that uses the variable.
- ## Step 4: give a verdict
- End with one of three lines: safe to publish, safe after these fixes (list them), or do not publish yet (say why). Keep it short.
- ## What this is not
- It is a first pass, not an audit. It will miss secrets with unusual shapes and it cannot see inside a hosted database or a vendor's systems. If the project handles customer data, payments or health information, get a professional review.
README.md · README · 43 lines
- # ship-check
- A Claude Code skill that checks a project for leaked passwords, API keys and tokens before you publish it. Built for people who make things with AI and are not security engineers.
- It started with one finding: when you click "don't ask again" on a command in Claude Code, the whole command is saved to `.claude/settings.local.json`. If you typed a key into that command, the key is in the file. Plenty of people have published that file without knowing.
- ## What it does
- 1. Runs a scan of the folder and its git history (`scripts/ship-check.sh`).
- 2. Explains each finding in plain words and tells you what to do about it.
- 3. Walks you through four checks a script cannot do: open the app logged out, check your share links, list what your AI agent can write to, and stop approving commands that have passwords in them.
- It never prints a secret. You see the file, the line and the first four characters.
- ## Install
- 1. Download this repo (Code, then Download ZIP) and unzip it.
- 2. Move the `ship-check` folder into your Claude Code skills folder: `~/.claude/skills/`
- 3. Open Claude Code in any project and say: `run a ship check on this folder`
- No Claude Code? The script runs on its own:
- ```bash
- bash scripts/ship-check.sh /path/to/your/project
- ```
- It needs bash, grep, awk and find, which every Mac and Linux machine has. git is optional.
- ## Example
- `examples/sample-report.txt` is a real run on a practice folder with made-up keys. It found a saved command with a token in it, a password in a config file, the same password in git history, and a `.env` file that was about to ship.
- ## What it will not catch
- Secrets with unusual shapes, anything inside a hosted database, and anything on a vendor's side. It is a first pass, not an audit. If your project handles customer data or payments, get a professional review.
- ## If it finds a real key
- Rotate it first. Make a new key at the provider and turn the old one off. Clean up the files second. A key that was public for a minute should be treated as stolen.
- ## License
- MIT. Use it, change it, share it.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.