secret-safe

What This Skill Does

The secret-safe skill is a foundational security component for OpenClaw agents, designed to eliminate the common vulnerability of exposing credentials within the LLM context. By enforcing strict environment-based injection patterns, this skill ensures that API keys, tokens, and passwords never touch the agent's reasoning process, history logs, or output artifacts. It provides standardized methodologies for handling sensitive authentication data, ensuring they remain encrypted in transit and at rest within the host environment rather than being stored in human-readable prompt history.

Installation

To integrate this protective layer, run the following command in your terminal: clawhub install openclaw/skills/skills/brycexbt/secret-safe Once installed, you should update your openclaw.json configuration to map your credentials to specific environment variables, ensuring the metadata frontmatter of your skill correctly defines the requires.env gate. This prevents any skill from executing unless the required security context is initialized.

Use Cases

• Automated API Integration: Securing connections to services like GitHub, AWS, or OpenAI without pasting raw keys into the chat interface.
• Security Audits: Scanning existing skills or developer-written instructions to identify and remediate potential credential leaks.
• Enterprise Configuration: Implementing secrets manager workflows (like HashiCorp Vault, 1Password, or Keychain) where the agent fetches secrets directly into the subprocess environment, bypassing the LLM layer entirely.
• Policy Enforcement: Helping developers structure their SKILL.md files to adhere to OpenClaw's security-first architectural standards.

Example Prompts

1. "How can I securely pass my Stripe API key to the new payment-processing skill I am building?"
2. "Audit my current SKILL.md file for credential exposure and suggest the correct environment injection pattern."
3. "Configure my agent to use the 1Password CLI to fetch my database credentials at runtime instead of hardcoding them."

Tips & Limitations

• Strict Isolation: Always ensure you are using double quotes around variables in shell commands to prevent command expansion leaks.
• Audit Regularly: While this skill helps mitigate risks, developers must still perform periodic manual reviews of their subprocess calls to ensure no accidental logging occurs.
• Environment Only: The skill only functions correctly if the agent adheres to the principle that the LLM should never be told the value of a secret, only that a secret exists within its environment.