Summary

Add a "masked secrets" system that allows agents to use API keys without being able to see them. This prevents accidental leaks and protects against prompt injection attacks designed to extract credentials.

Problem

Currently, secrets stored in ~/.openclaw/.env are fully accessible to agents:

1. Agent can run cat ~/.openclaw/.env and read plaintext keys
2. Agent can run echo $API_KEY to print environment variables
3. Secrets can appear in command outputs and chat history
4. Prompt injection attacks could trick an agent into revealing secrets

Proposed Solution

1. Secret Reference Syntax

curl -H "Authorization: Bearer {{secret:MY_API_KEY}}" https://api.example.com

2. Runtime Substitution — Gateway substitutes the real value before execution

3. Output Redaction — Any leaked secret replaced with [REDACTED]

4. Blocked Commands — Prevent echo $VAR, env, cat .env

5. Config:

secrets:
  mask:
    - MY_API_KEY
    - ANOTHER_SECRET

Willingness to Contribute

Happy to implement this if the approach looks good!

This proposal was developed with AI assistance (Claude/OpenClaw).