fix(agent): guard openai parse_response against null response.output

[marcelopaniza]

Why

0.14.0 crashes on chatgpt.com/backend-api/codex with TypeError: 'NoneType' object is not iterable when the backend returns response.output: null. Tracked in #33578.

agent/codex_runtime.py on main is already structurally immune — it never reads response.completed.response.output for content. But any caller still on the high-level responses.stream(...) / responses.parse(...) helpers (third-party plugins, future provider profiles, regressions) remains exposed to the same SDK bug as long as openai==2.24.0 is pinned.

What

• agent/_openai_compat.py — wraps openai.lib._parsing._responses.parse_response. If the incoming response.output is None, it's coerced to [] before delegating to the original function. The wrapper re-binds the symbol in the two SDK modules that imported it by name (openai.lib.streaming.responses._responses, openai.resources.responses.responses) so already-loaded code paths pick up the guard.
• agent/__init__.py — imports _openai_compat so installation happens on first agent use, before any provider profile is exercised.
• tests/agent/test_openai_compat.py — three tests: module importable, guard active on the canonical symbol, streaming/resource modules use the guarded reference.

Scope

• Idempotent — second install is a no-op (guarded by _hermes_or_empty_guard flag on the function).
• No-op once upstream openai-python ships the guard (fix(parsing): handle null response.output in parse_response openai/openai-python#3322 / fix(signal): track SSE keepalive comments as connection activity #3316 / fix(terminal): preserve partial output when command times out #3286). The wrapper coerces None → [], which the fixed parser also accepts.
• Does not touch agent/codex_runtime.py — main's structural fix stays the primary defense.

Test

pytest tests/agent/test_openai_compat.py \
       tests/agent/test_auxiliary_client.py \
       tests/run_agent/test_codex_xai_oauth_recovery.py \
       tests/run_agent/test_streaming.py

→ 263 passed locally (Python 3.14.4 / openai==2.24.0).

Refs: #33578, openai/openai-python#3322, openai/openai-python#3316, openai/openai-python#3286.

🤖 Generated with Claude Code

[teknium1]

Thanks for this — closing as already fixed on main.

The Codex null-output crash (response.completed with output: null from the chatgpt.com backend → TypeError: 'NoneType' object is not iterable → surfaced as "non-retryable HTTP None") is resolved by commits already on main:

• cb38ce28c — drop the SDK responses.stream() helper and consume events directly, rebuilding output from response.output_item.done events so the null terminal-event output is never iterated (refactor(codex): drop SDK responses.stream() helper; consume events directly #33042).
• 43a3f119f — recover Codex streams with null output.
• dc9d677d5 — classify TypeError('NoneType … not iterable') as a retryable provider-shape error so retry/fallback runs instead of killing the turn.

Your fix targets the same path that's now hardened upstream, so there's nothing left to merge here. Your authorship is preserved in your branch; I'm crediting your report in the umbrella issue #33932 (now closed). Appreciate the contribution.