$ the-wire · showcase
Codex subagents inherit root service tier; openai-python releases 3.6.0
By RepoJournal · Filed · About OpenAI
Codex subagents now follow the root thread's service tier, and openai-python 3.6.0 adds compute_units to usage metrics.
Codex subagents now inherit the root thread's current service tier, applying it to each subagent request when the selected child model supports it, including remote compaction requests [1]. This removes per-spawn and role-level service-tier overrides so routing is controlled by the root thread. In the same repo, history item lookups get lazy indexing for larger thread-history turns, and cloud task credentials are now validated against trusted ChatGPT HTTPS origins on port 443, rejecting URLs with user info, queries, or fragments [2]. MCP HTTP helpers re-run after 401 or 403 responses and retry once when headers change, and runtimes refresh on remote plugin state changes [3][4]. openai-node prevents Responses WebSockets from following redirects even when followRedirects: true is set, surfacing the first response as an error like "Unexpected server response: 302" [5]. openai-python 3.6.0 adds compute_units to Responses and Chat Completions usage, and its embeddings parser now checks NumPy availability once per response instead of per embedding, cutting repeated checks [6][7].
Action items
- → Update Codex deployments to benefit from root-tier enforcement and the cloud credential origin validation openai/codex [plan]
- → Pin openai-python to 3.6.0 to get compute_units in usage payloads openai/openai-python [plan]
- → Review openai-node Responses WebSocket callers for behavior changes when redirects were previously allowed openai/openai-node [immediate]
References
- [1] Make subagents follow the root service tier (#41308) ↗ openai/codex
- [2] Restrict cloud task credentials to trusted origins ↗ openai/codex
- [3] Refresh MCP HTTP helper headers after authorization failures ↗ openai/codex
- [4] Refresh runtimes for remote plugin state changes ↗ openai/codex
- [5] fix(api): prevent Responses WebSockets from following redirects (#2513) ↗ openai/openai-node
- [6] v3.6.0 ↗ openai/openai-python
- [7] fix: avoid repeated numpy checks for embeddings ↗ openai/openai-python