hol-guard

Why we included this project

When coding agents like Claude Code, Codex, Cursor, or Gemini CLI run on shared machines, one bad tool call can exfiltrate keys or install a malicious package. HOL Guard handles that by sitting between the agent and its tools, inspecting risky operations at runtime: it flags prompt injection, blocks access to secrets, stops bad package installs, and checks MCP servers, plugins, and skills before they execute. Because it runs locally and ships with a separate plugin-scanner package, maintainers can also vet packages inside CI rather than relying on the agent to behave. Teams with real supply-chain or data-leakage concerns will find the blocking behavior quick to test against their own agent workflows.