Mesh Governance System

Policy enforcement layer deployed identically across every node in the mesh. Behavioural hooks intercept tool calls before execution, evaluating each action against a rule set: protected path enforcement (files/directories that cannot be deleted), destructive command detection (force pushes, hard resets, recursive deletes), backup age validation (prevents restoring stale backups without confirmation), large file warnings, wildcard expansion confirmation, and process kill safety checks.

Every tool invocation is logged to an immutable audit trail with timestamp, node, session, tool name, action attempted, decision outcome (allowed/blocked/warned), rule evaluated, and full context. Blocked actions trigger real-time desktop notifications on the control node. The audit trail is queryable by node, action type, decision, and time range — with 24-hour aggregate statistics showing block rates by rule and by node.

Pre-flight validation API allows services to check whether an action would be permitted before attempting it. Protected path registry supports pattern matching with configurable protection levels and documented reasons. The governance rules are the same on every node — identical hook scripts, identical rule evaluation, consistent policy enforcement regardless of which machine an AI instance is running on.

Hook system also handles session lifecycle: session start fetches unread inter-node messages and active session summaries, periodic checks during sessions surface new messages, and session end auto-posts a summary to the messaging system. This integrates governance with operational awareness — the same hook infrastructure that enforces safety rules also keeps instances informed of cross-node activity.