Squelette – keep AI coding agents inside the scope you approved (github.com)

Squelette is a new governance framework designed to maintain control over AI-assisted projects by clearly defining objectives, scopes, and human authority throughout the development process. This system addresses the common issue of project drift, where AI agents may forget constraints or exceed their permissions. Squelette operates through a controller that imposes strict checks at each stage, ensuring that only authorized work items are executed, and requires evidence of completion before declaring a project as done. The significance of Squelette lies in its ability to enhance project accountability and traceability in environments where multiple AI agents collaborate over extended periods. By utilizing standard-library Python for its operations, Squelette minimizes dependencies and facilitates consistent governance regardless of the AI tools in use. The framework promotes a clear record-keeping system using Git provenance, ensuring that every decision and action aligns with human-defined parameters. This structured approach not only safeguards against deviations but also simplifies project management, allowing for the seamless integration of AI agents while providing a transparent audit trail for human teams.