The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to clone source code from a repository at github.com/O-ozzz/OctoGuard--Free-OpenClaw-Security-Supervision-System.git, which is an unverified third-party source.
[REMOTE_CODE_EXECUTION]: The skill executes the downloaded content using npm install and npm start, allowing for the execution of code from the remote repository during the setup process.
[PERSISTENCE]: Deployment steps include the use of pm2 startup and pm2 save, which configures the security backend to persist across system reboots as a background service.
[INDIRECT_PROMPT_INJECTION]: The skill functions as a transparent proxy for AI agent conversations and tool executions, introducing a vulnerability surface where data from user inputs or tool outputs is processed for security auditing.
Ingestion points: Intercepts conversation data and tool execution results through the /api/proxy endpoint and WebSocket monitoring stream.
Boundary markers: No explicit delimiter or "ignore embedded instructions" warnings are defined in the proxy configuration.
Capability inventory: The system monitors file access (e.g., /etc/passwd), command execution (e.g., rm -rf), and performs network-based alert notifications.
Sanitization: Uses regular expression patterns for matching high-risk behaviors but does not specify further input validation or escaping for data passing through the proxy.

octoguard-openclaw-security-governance