Show HN: Scanning AI agents for dangerous permissions and a security layer

MakerChecker is an open-source security layer compatible with existing frameworks such as LangChain, Claude SDK, and CrewAI. It prevents agents from calling tools they are not allowed to use, and leaves a cryptographically signed audit log of every execution.

Key features include:

• Code scanning (mc scan): Scans the actions an agent can perform, classifies them by risk level, and automatically generates security policy code.
• Permission control (embedded): Applies a 'Deny-by-default' principle, restricting execution to only the Skills granted to predefined Roles.
• Audit and approval: Designed so that an agent cannot approve its own work, with every decision recorded in a tamper-proof log via Ed25519 signatures and Hash-chaining technology.

Users can run a self-hosted server via Docker to build a human-in-the-loop approval process and review console.