Agent Skills: Varlock Security Skill

Varlock Security Skill

Secure-by-default environment variable management for Claude Code sessions.

Repository: https://github.com/dmno-dev/varlock Documentation: https://varlock.dev

Core Principle: Secrets Never Exposed

When working with Claude, secrets must NEVER appear in:

• Terminal output
• Claude's input/output context
• Log files or traces
• Git commits or diffs
• Error messages

This skill ensures all sensitive data is properly protected.

CRITICAL: Security Rules for Claude

Rule 1: Never Echo Secrets

# ❌ NEVER DO THIS - exposes secret to Claude's context
echo $CLERK_SECRET_KEY
cat .env | grep SECRET
printenv | grep API

# ✅ DO THIS - validates without exposing
varlock load --quiet && echo "✓ Secrets validated"

Rule 2: Never Read .env Directly

# ❌ NEVER DO THIS - exposes all secrets
cat .env
less .env
Read tool on .env file

# ✅ DO THIS - read schema (safe) not values
cat .env.schema
varlock load  # Shows masked values

Rule 3: Use Varlock for Validation

# ❌ NEVER DO THIS - exposes secret in error
test -n "$API_KEY" && echo "Key: $API_KEY"

# ✅ DO THIS - Varlock validates and masks
varlock load
# Output shows: API_KEY 🔐sensitive └ ▒▒▒▒▒

Rule 4: Never Include Secrets in Commands

# ❌ NEVER DO THIS - secret in command history
curl -H "Authorization: Bearer sk_live_xxx" https://api.example.com

# ✅ DO THIS - use environment variable
curl -H "Authorization: Bearer $API_KEY" https://api.example.com
# Or better: varlock run -- curl ...

Quick Start

Installation

# Install Varlock CLI
curl -sSfL https://varlock.dev/install.sh | sh -s -- --force-no-brew

# Add to PATH (add to ~/.zshrc or ~/.bashrc)
export PATH="$HOME/.varlock/bin:$PATH"

# Verify
varlock --version

Initialize Project

# Create .env.schema from existing .env
varlock init

# Or create manually
touch .env.schema

Schema File: .env.schema

The schema defines types, validation, and sensitivity for each variable.

Basic Structure

# Global defaults
# @defaultSensitive=true @defaultRequired=infer

# Application
# @type=enum(development,staging,production) @sensitive=false
NODE_ENV=development

# @type=port @sensitive=false
PORT=3000

# Database - SENSITIVE
# @type=url @required
DATABASE_URL=

# @type=string @required @sensitive
DATABASE_PASSWORD=

# API Keys - SENSITIVE
# @type=string(startsWith=sk_) @required @sensitive
STRIPE_SECRET_KEY=

# @type=string(startsWith=pk_) @sensitive=false
STRIPE_PUBLISHABLE_KEY=

Security Annotations

| Annotation | Effect | Use For | |------------|--------|---------| | @sensitive | Redacted in all output | API keys, passwords, tokens | | @sensitive=false | Shown in logs | Public keys, non-secret config | | @defaultSensitive=true | All vars sensitive by default | High-security projects |

Type Annotations

| Type | Validates | Example | |------|-----------|---------| | string | Any string | @type=string | | string(startsWith=X) | Prefix validation | @type=string(startsWith=sk_) | | string(contains=X) | Substring validation | @type=string(contains=+clerk_test) | | url | Valid URL | @type=url | | port | 1-65535 | @type=port | | boolean | true/false | @type=boolean | | enum(a,b,c) | One of values | @type=enum(dev,prod) |

Safe Commands for Claude

Validating Environment

# Check all variables (safe - masks sensitive values)
varlock load

# Quiet mode (no output on success)
varlock load --quiet

# Check specific environment
varlock load --env=production

Running Commands with Secrets

# Inject validated env into command
varlock run -- npm start
varlock run -- node script.js
varlock run -- pytest

# Secrets are available to the command but never printed

Checking Schema (Safe)

# Schema is safe to read - contains no values
cat .env.schema

# List expected variables
grep "^[A-Z]" .env.schema

Common Patterns

Pattern 1: Validate Before Operations

# Always validate environment first
varlock load --quiet || {
  echo "❌ Environment validation failed"
  exit 1
}

# Then proceed with operation
npm run build

Pattern 2: Safe Secret Rotation

# 1. Update secret in external source (1Password, AWS, etc.)
# 2. Update .env file manually (don't use Claude for this)
# 3. Validate new value works
varlock load

# 4. If using GitHub Secrets, sync (values not shown)
./scripts/update-github-secrets.sh

Pattern 3: CI/CD Integration

# GitHub Actions - secrets from GitHub Secrets
- name: Validate environment
  env:
    DATABASE_URL: ${{ secrets.DATABASE_URL }}
    API_KEY: ${{ secrets.API_KEY }}
  run: varlock load --quiet

Pattern 4: Docker Integration

# Install Varlock in container
RUN curl -sSfL https://varlock.dev/install.sh | sh -s -- --force-no-brew \
    && ln -s /root/.varlock/bin/varlock /usr/local/bin/varlock

# Validate at container start
CMD ["varlock", "run", "--", "npm", "start"]

Handling Secret-Related Tasks

When User Asks to "Check if API key is set"

# ✅ Safe approach
varlock load 2>&1 | grep "API_KEY"
# Shows: ✅ API_KEY 🔐sensitive └ ▒▒▒▒▒

# ❌ Never do
echo $API_KEY

When User Asks to "Debug authentication"

# ✅ Safe approach - check presence and format
varlock load  # Validates types and required fields

# Check if key has correct prefix (without showing value)
varlock load 2>&1 | grep -E "(CLERK|AUTH)"

# ❌ Never do
printenv | grep KEY

When User Asks to "Update a secret"

Claude should respond:
"I cannot directly modify secrets for security reasons. Please:
1. Update the value in your .env file manually
2. Or update in your secrets manager (1Password, AWS, etc.)
3. Then run `varlock load` to validate

I can help you update the .env.schema if you need to add new variables."

When User Asks to "Show me the .env file"

Claude should respond:
"I won't read .env files directly as they contain secrets. Instead:
- Run `varlock load` to see masked values
- Run `cat .env.schema` to see the schema (safe)
- I can help you modify .env.schema if needed"

External Secret Sources

1Password Integration

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

AWS Secrets Manager

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

Environment-Specific Values

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

Troubleshooting

"varlock: command not found"

# Check installation
ls ~/.varlock/bin/varlock

# Add to PATH
export PATH="$HOME/.varlock/bin:$PATH"

# Or use full path
~/.varlock/bin/varlock load

"Schema validation failed"

# Check which variables are missing/invalid
varlock load  # Shows detailed errors

# Common fixes:
# - Add missing required variables to .env
# - Fix type mismatches (port must be number)
# - Check string prefixes match schema

"Sensitive value exposed in logs"

# 1. Rotate the exposed secret immediately
# 2. Check .env.schema has @sensitive annotation
# 3. Ensure using varlock commands, not echo/cat

# Add missing sensitivity:
# Before: API_KEY=
# After:  # @type=string @sensitive
#         API_KEY=

npm Scripts

Add these to your package.json:

{
  "scripts": {
    "env:validate": "varlock load",
    "env:check": "varlock load --quiet || echo 'Environment validation failed'",
    "prestart": "varlock load --quiet",
    "start": "varlock run -- node server.js"
  }
}

Security Checklist for New Projects

• [ ] Install Varlock CLI
• [ ] Create .env.schema with all variables defined
• [ ] Mark all secrets with @sensitive annotation
• [ ] Add @defaultSensitive=true to schema header
• [ ] Add .env to .gitignore
• [ ] Commit .env.schema to version control
• [ ] Add npm run env:validate to CI/CD
• [ ] Document secret rotation procedure
• [ ] Never use cat .env or echo $SECRET in Claude sessions

Quick Reference Card

| Task | Safe Command | |------|-------------| | Validate all env vars | varlock load | | Quiet validation | varlock load --quiet | | Run with env | varlock run -- <cmd> | | View schema | cat .env.schema | | Check specific var | varlock load \| grep VAR_NAME |

| Never Do | Why | |----------|-----| | cat .env | Exposes all secrets | | echo $SECRET | Exposes to Claude context | | printenv \| grep | Exposes matching secrets | | Read .env with tools | Secrets in Claude's context | | Hardcode in commands | In shell history |

Integration with Other Skills

Clerk Skill

• Test user passwords are @sensitive
• Test emails are @sensitive=false (contain +clerk_test, not secret)
• See: ~/.claude/skills/clerk/SKILL.md

Docker Skill

• Mount .env file, never copy secrets to image
• Use varlock run as entrypoint
• See: ~/.claude/skills/docker/SKILL.md

Last updated: December 22, 2025 Secure-by-default environment management for Claude Code