You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Linux 5.15.167.4-microsoft-standard-WSL2 x86_64 x86_64
What issue are you seeing?
In the VS Code Codex runtime, spawned explorer subagents can ignore the explicit model
override passed in the spawn_agent call.
The parent session requests model="gpt-5.4-mini", but the spawned child thread starts with turn_context.model = "gpt-5.4".
This makes repo-level subagent configs such as .codex/agents/context-gatherer.toml ineffective
for explorer agents even when the parent agent forwards the configured model explicitly.
What steps can reproduce the bug?
Start a parent Codex session in VS Code with default model gpt-5.4.
From the parent, spawn a read-only explorer subagent with:
agent_type="explorer"
model="gpt-5.4-mini"
reasoning_effort="medium"
fork_context=true
Inspect the spawned child session log.
What is the expected behavior?
The child session should start with turn_context.model = "gpt-5.4-mini".
Additional information
This repro focuses on explorer subagents because the failure is directly confirmed there.
Repo guidance was already updated to explicitly pass model="gpt-5.4-mini" for context-gatherer spawns, so this appears to be a runtime issue rather than a repo-config issue.
The mismatch is visible in persisted session logs even when the parent call arguments are correct.
Activity
github-actions commentedon Mar 22, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
zibb0 commentedon Mar 22, 2026
closed as duplicate