snoopy

snoopy

I found that sometimes Claude Code reads whatever it wants. It'll open my .env, scan my .pem files, grep through my credentials.

Once a secret hits the context window, it's in the prompt being sent to Anthropic's servers and I am not a huge fant of that. I like control (that sounds crazy, but you know what I mean)

Load the plugin when starting Claude:

claude --plugin-dir /path/to/snoopy

How it works

Snoopy installs a PreToolUse hook that fires before every tool call. If Claude tries to read, edit, search, or glob a file that matches a secret pattern, the operation gets denied.

What gets blocked

Files:

• .env, .env.*, .env.local, .env.production
• *.pem, *.key
• secrets.json, credentials.*
• *.secret, *.keystore

Bash commands:

• cat .env, head credentials.json, tail secrets.key, etc.
• echo $SECRET_*, echo $API_KEY_*, printenv

Managing patterns

Type /guard to open the interactive picker. It scans the project for sensitive files and lets you block, allow, or remove patterns you want!

You can also be boring and edit ~/.snoopy/config.json directly:

{
  "blocked": [".env", ".env.*", "*.pem", "*.key", "secrets.json", "credentials.*", "*.secret", "*.keystore"],
  "allowed": [".env.example", ".env.template", ".env.sample"]
}