Secret Redactor

An OpenCode plugin that prevents secrets from leaking into LLM context. Secrets detected in tool output are replaced with 🔒label🔓 tokens before the model sees them, then transparently restored when a tool needs the real value for execution.

Detected secret types

AWS keys, GitHub/GitLab tokens, OpenAI/Anthropic keys, Google Cloud credentials, Stripe keys, Slack tokens, JWTs, private keys, database connection strings, and many more. See src/patterns.ts for the full list.

Setup

Add the plugin to your opencode.json:

{
  "plugin": ["opencode-secret-redactor@0.5.1"]
}

The plugin hooks into tool execution automatically -- no further configuration is required.

How it works

1. After bash or read tool output, the plugin scans for secrets using pattern matching and stores any matches in an in-memory vault.
2. The output sent to the LLM contains only redacted placeholders.
3. Before bash, write, or edit tool execution, placeholders in the tool arguments are replaced with the original values so commands run correctly.

License

MIT