Codex¶
Source: openai/codex
Writes ~/.codex/hooks.json (SessionStart, UserPromptSubmit, Stop, SessionEnd), [mcp_servers.murmur] in ~/.codex/config.toml, and ~/.codex/prompts/murmur.md.
Codex asks you to trust new hooks once: open /hooks. The slash command is /prompts:murmur.
How it connects¶
Hooks run murmur hook <event> --client codex and receive session_id and transcript_path. Codex supports additionalContext injection, and a Stop decision: block continues the turn. MCP tools come from [mcp_servers.murmur].
Delivery path¶
- Live push:
codex queue --thread <id>; a running window takes it as a user turn. The push counts assentwhencodex queueexits 0. - Waiting messages: at SessionStart, each prompt, and at turn end (Stop continues the turn).
- Evidence for
delivered: the tag in the receiver's rollout transcript (JSONL).
Known limits¶
- New hooks need the one-time trust step in
/hooks. - Researched against Codex 0.154.0.