MakerChecker
An open source security layer for AI agents that enforces role-based access control, requires human approval for high-risk actions, and writes a cryptographically signed, tamper-evident audit log.
MakerChecker sits in front of every tool call an agent makes as a checkpoint, and behind it as a signed ledger, working alongside existing frameworks such as LangChain, the Claude SDK or CrewAI. An agent acts only through an assigned role, can run only the skills it has been granted, cannot exceed its limits, and is denied before execution for anything outside its role.
A scanning command inspects a codebase for what an agent could already do unsupervised, classifies each action by risk and can generate the governance code to fix it. An embeddable package wraps any tool call with these controls, routing high-risk skills to a separate role so an agent can never approve its own actions, and a self-hosted server adds a durable, queryable audit trail, a human-approval inbox and a review console, with every decision Ed25519-signed and hash-chained so a changed row breaks verification.