Context
Codex's docs describe subagent-related hooks (subagent hooks report the parent's session_id, per research done during Map: Harness-grounded graph model across the Context Graph family), but agent-context-graph's codex.py adapter never implemented them — _SUPPORTED_HOOKS only covers SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, PermissionRequest, Stop. No AgentStartEvent/AgentEndEvent construction exists for Codex at all, so no (:Agent) node (map #275, tickets #277/#278) is ever created for a Codex session today, regardless of whether Codex itself spawns something analogous to a subagent.
Surfaced while scoping a new map for live-harness verification of the SPAWNED subagent-nesting inference rule — verifying against a real Codex session turned out to be blocked on this gap, and was ruled out of scope for that map rather than bundled in, since this is real adapter-design work (what does Codex's actual subagent hook payload look like? does its agent_id concept work the same way as Claude Code's?) rather than a testing-methodology decision.
Scope
- Determine Codex's actual subagent-hook payload shape (may need its own research pass against current Codex docs/CLI, not assumed from memory).
- Wire
SubagentStart/SubagentStop-equivalent hooks into codex.py, mirroring claude_code.py's AgentStartEvent/AgentEndEvent construction.
- Likely candidate for its own wayfinder map, given the open design questions above.
Context
Codex's docs describe subagent-related hooks (subagent hooks report the parent's session_id, per research done during Map: Harness-grounded graph model across the Context Graph family), but
agent-context-graph'scodex.pyadapter never implemented them —_SUPPORTED_HOOKSonly coversSessionStart,UserPromptSubmit,PreToolUse,PostToolUse,PermissionRequest,Stop. NoAgentStartEvent/AgentEndEventconstruction exists for Codex at all, so no(:Agent)node (map #275, tickets #277/#278) is ever created for a Codex session today, regardless of whether Codex itself spawns something analogous to a subagent.Surfaced while scoping a new map for live-harness verification of the
SPAWNEDsubagent-nesting inference rule — verifying against a real Codex session turned out to be blocked on this gap, and was ruled out of scope for that map rather than bundled in, since this is real adapter-design work (what does Codex's actual subagent hook payload look like? does itsagent_idconcept work the same way as Claude Code's?) rather than a testing-methodology decision.Scope
SubagentStart/SubagentStop-equivalent hooks intocodex.py, mirroringclaude_code.py'sAgentStartEvent/AgentEndEventconstruction.