A guard-railed, closed-loop workflow for AI coding agents: live state bus + execution-level hard intercepts for Claude Code and Codex (GitHub PR / GitLab MR). From step-level to requirement-level; eval-driven, spec-driven, human-in-the-loop.
Hook
Runs before the agent uses a tool for Bash, Edit, Write and NotebookEdit tool calls, executing tool_call_timeline.py, pretool_policy_bash.py and pretool_policy_edit.py (3 commands). From compforge/devloop.
Hook
Runs when a session starts, executing sessionstart_init.py. From compforge/devloop.
Hook
Runs when a session ends, executing sessionend_release.py. From compforge/devloop.
Hook
Runs when you submit a prompt, before the agent sees it, executing userprompt_inject.py. From compforge/devloop.
Hook
Runs after the context is compacted, executing postcompact_reinject.py. From compforge/devloop.
Hook
Runs on the FileChanged event, executing filechanged_refs.py. From compforge/devloop.
Hook
Runs on the CwdChanged event, executing cwdchanged_enter.py. From compforge/devloop.
Hook
Runs after a tool call finishes for Bash tool calls, executing tool_call_timeline.py and posttool_git_refresh.py (2 commands). From compforge/devloop.
Hook
Runs when a tool call fails, executing tool_call_timeline.py. From compforge/devloop.