fix: guard against None message in structured_output_response

[qizwiz]

What

Fix two places in augmented_llm_openai.py where the existing guard

if not completion.choices or completion.choices[0].message.content is None:

is itself vulnerable to AttributeError when message is None.

Why

If completion.choices[0].message is None (e.g. Gemini via OpenAI-compatible endpoint on PROHIBITED_CONTENT, which returns HTTP 200 with message: null), accessing .content in the condition itself throws AttributeError: 'NoneType' object has no attribute 'content' — before the guard can fire.

Fix

# Before (crashes when message is None):
if not completion.choices or completion.choices[0].message.content is None:
    raise ValueError("No structured content returned by model")

# After (safe):
if not completion.choices or completion.choices[0].message is None:
    raise ValueError("No structured content returned by model")
content = completion.choices[0].message.content
if content is None:
    raise ValueError("No structured content returned by model")

Both the sync and async structured_output_response paths are fixed.
No behaviour change for well-formed responses.

Summary by CodeRabbit

• Bug Fixes
  • Strengthened validation of AI model responses to more reliably detect incomplete or missing data
  • Enhanced error messages for structured output generation failures, providing clearer feedback when the model cannot generate the expected output

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 570904be-df8a-4544-b4d0-75a306d38ab2

📥 Commits

Reviewing files that changed from the base of the PR and between f62d849 and d6ba7d0.

📒 Files selected for processing (1)

• src/mcp_agent/workflows/llm/augmented_llm_openai.py

---

📝 Walkthrough

Walkthrough

The PR tightens structured-output validation in OpenAI response handlers by splitting combined null checks into granular steps. Two methods—generate_structured and request_structured_completion_task—now explicitly validate completion.choices existence, first choice message presence, and message.content non-nullity separately, raising ValueError with consistent messaging when any check fails.

Changes

Validation Tightening for Structured OpenAI Responses

Layer / File(s)	Summary
Granular validation in OpenAI response handlers src/mcp_agent/workflows/llm/augmented_llm_openai.py	Both generate_structured and request_structured_completion_task now validate completion.choices and first choice message separately before reading message.content, raising ValueError with consistent messaging when any step fails.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Suggested reviewers

• saqadri
• rholinshead

Poem

🐰 Oh, what validation precision fine,
Each choice and message we divine,
Split checks that guard the model's reply,
With structured care, no stone left dry! 🌿

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main change: adding guards against None message in structured output response handling.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}