llm_chat

A guardrails system that lets a Claude Code session run unattended safely. Provides autonomy engine (watchdog + stopgate) and guardrails (claim, write guard, arm->consume primitive) to keep AI sessions alive and prevent unsafe mutations.