I was trying to figure out why my /ce-code-review calls kept always responding with something along the lines of
I emulated the CE reviewer lenses in the main thread because this repo’s AGENTS.md maps subagent dispatch to main-thread review.
and realized this line is added by the CE compatibility layer:
- Task (subagent dispatch) / Subagent / Parallel: run sequentially in main thread; use multi_tool_use.parallel for tool calls
which makes me wonder - did you guys find that Codex behaves better given that instruction or was that potentially an oversight from a previous version (before subagents?) that never got addressed?
Personally I've found that if I instruct the agent to ignore that instruction and dispatch the reviewer personas as subagents anyway it tends to find more issues in code, though I haven't yet been able to fully discern whether there are more false positives or just more real findings.
I was trying to figure out why my
/ce-code-reviewcalls kept always responding with something along the lines ofand realized this line is added by the CE compatibility layer:
- Task (subagent dispatch) / Subagent / Parallel: run sequentially in main thread; use multi_tool_use.parallel for tool callswhich makes me wonder - did you guys find that Codex behaves better given that instruction or was that potentially an oversight from a previous version (before subagents?) that never got addressed?
Personally I've found that if I instruct the agent to ignore that instruction and dispatch the reviewer personas as subagents anyway it tends to find more issues in code, though I haven't yet been able to fully discern whether there are more false positives or just more real findings.