Varlock Skill

What it is

Varlock Skill is William Smith’s one-file skill that teaches Claude Code to handle secrets without reading them. It wraps Varlock, a DMNO tool that reads a .env.schema file beside your .env and masks each value marked sensitive. With the skill on, Claude checks that a key is set by running varlock load. It reads the schema file in place of .env.

1. You install the Varlock tool and the skill file
2. You write .env.schema, with no secret values in it
3. You ask Claude about a key or a sign-in problem
4. Claude runs varlock load and never opens .env
5. Varlock prints each secret masked

Who it suits

Developers who use Claude Code in a project with API keys, tokens or database addresses in a .env file and do not want those values in a chat. It suits small teams too, since the schema file holds no secret values and can be committed.

Where to use it

Claude Code is the home. The README is titled for it and names no other agent.

• Claude Code, every project: the skill file at ~/.claude/skills/varlock/SKILL.md
• Any project with a .env: the Varlock tool on your PATH and a .env.schema in the project
• GitHub Actions: a step that runs varlock load --quiet
• Docker: Varlock installed in the image, and varlock run as the container’s start command

How to set it up

You install two things. The Varlock tool does the masking, and the skill file tells Claude when to use it. Get it saves a zip of Mindket’s setup notes in your downloads folder.

1. Install the Varlock tool. You run the installer from the README, then put its folder on your PATH.
2. Add the skill file. One line makes ~/.claude/skills/varlock and saves SKILL.md in it.
3. Write a schema for your project. Run varlock init to build .env.schema from your .env, or write it by hand.
4. Check the setup. Run varlock --version, then varlock load. In Claude Code, /skills should list varlock.

curl -sSfL https://varlock.dev/install.sh | sh -s -- --force-no-brew
export PATH="$HOME/.varlock/bin:$PATH"

The first line runs Varlock’s install script the moment it downloads, so read Good to know before you paste it. The skill file says to add the export line to ~/.zshrc or ~/.bashrc.

mkdir -p ~/.claude/skills/varlock && curl -sSL https://raw.githubusercontent.com/wrsmith108/varlock-claude-skill/main/skills/varlock/SKILL.md -o ~/.claude/skills/varlock/SKILL.md

The README’s second route clones the repository to /tmp, copies skills/varlock into ~/.claude/skills/ and deletes the clone. Make ~/.claude/skills first with mkdir -p. On a machine that has none, cp -r would name the new folder skills and leave SKILL.md one level too high.

How to use it

Ask Claude about your keys in plain words. With the skill loaded, Claude reaches for varlock in place of echo, cat or printenv.

Example

Payments fail in your app, and you suspect the Stripe key is missing from .env.

Is my Stripe key set? Do not show me the value.

Claude runs varlock load, finds STRIPE_SECRET_KEY marked sensitive with its value masked, and tells you it is set.

If the line says the variable is missing, Claude says so and asks you to add the value to .env yourself. It will not write the value for you.

For newcomers and experts

If you are new. A .env file holds your keys as plain text. If Claude reads it, every key lands in the chat. Varlock splits the job in two. The schema file names each variable, its type and whether it is secret. It holds no secret values. The varlock load command checks the real .env and prints a masked answer. Start in a small project and run varlock init.

For experts. The schema opens with a header line, then one line of annotations above each variable. The skill file lists seven types: string, string(startsWith=X), string(contains=X), url, port, boolean and enum(a,b,c). A value can come from a command, written as exec(...) in the schema. These are the skill file’s lines for 1Password and AWS, and for a value picked by environment:

# @type=string @sensitive
API_KEY=exec('op read "op://vault/item/field"')

# @type=string @sensitive
DB_PASSWORD=exec('aws secretsmanager get-secret-value --secret-id prod/db')

# @type=url
API_URL=env('API_URL_${NODE_ENV}', 'http://localhost:3000')

The skill file’s GitHub Actions example passes the secrets in as env and runs varlock load --quiet. Its Docker example installs Varlock in the image and starts the app with varlock run. Its npm scripts example wires varlock load into a prestart script.

Good to know

The skill asks Claude to behave. It is not a lock.

• Enforcement: the repository holds one SKILL.md and a plugin manifest, with no hook or script that blocks a read of .env. The skill works only while Claude follows it
• The installer: the README’s Varlock line fetches https://varlock.dev/install.sh and pipes it straight into sh. The script runs at once, as your user, before you have read it. To read it first, open that address in a browser
• Windows: the README’s commands are written for bash and give no Windows steps
• Secrets are your job: Claude declines to change one and points you to your .env or secrets manager
• Loose ends: the skill file’s rotation pattern calls an update-github-secrets.sh script, and its last section points to a Clerk skill and a Docker skill. The repository holds none of them
• Maker: William Smith, MIT licensed, version 1.0.0, last updated December 22, 2025 by the file’s own note. Varlock itself is DMNO’s. A community project, not made or endorsed by Anthropic