agentguard

You can audit AI agent runtime actions for security vulnerabilities by scanning codebases and evaluating proposed network requests, file operations, and web3 transactions using configurable protection levels. This identifies malicious commands and prevents data leaks before execution.

AI agent runtime protection is a mechanism that evaluates proposed actions for safety before execution. It prevents secret exposure and data leaks by auditing code for command execution risks, remote loading, and unauthorized network access using the GoPlus AgentGuard framework.

To scan a newly discovered skill for malicious commands before execution, run a security audit on the current directory. This evaluates the skill's code for potential risks like command execution and secret exposure, then checks if it is safe to access your network.

Yes, agentguard supports web3 transaction safety checks. It evaluates proposed web3 transactions for safety risks alongside command execution, network requests, and file operations using configurable protection levels to manage skill trust and capabilities.

Yes, you can configure trust levels for different AI agent skills. The framework manages a registry of trusted skills and their capabilities, allowing you to set configurable protection levels for command execution, network access, and file operations.

Agentguard for codebase security auditing requires the @goplus/agentguard-cli dependency and focuses on command execution, remote loading, and secret exposure. It evaluates runtime actions but requires proper configuration of protection levels to avoid blocking legitimate agent workflows.