fix: guard None text in ItemHelpers.extract_last_content

[ioleksiuk]

Summary

Extends the fix in #3375 (`fix: guard None text in text_message_output ...`) to the two sibling helpers in `ItemHelpers` that have the same shape.

`ResponseOutputText.text` is typed as `str` per the Responses API schema, but `src/agents/items.py:714-720` already documents (in the comment above the original #3375 fix) that provider gateways like LiteLLM and `model_construct` paths during streaming surface `None` values. PR #3375 fixed `text_message_output` for this case. Two sibling helpers were missed:

• `extract_last_content` (declared `-> str`, items.py:678) — silently returned `None` when `last_content.text` was `None`, violating its declared type contract.
• `extract_last_text` (declared `-> str | None`, items.py:693) — returned `None` only by passthrough, so callers using truthy semantics couldn't distinguish "no text content" from "text is None".

Repro for `extract_last_content` (declared `-> str`, returns `None`):

```python

text_part = ResponseOutputText.model_construct(text=None, type="output_text", annotations=[])
msg = ResponseOutputMessage(id="m", role="assistant", status="completed",
... type="message", content=[text_part])
ItemHelpers.extract_last_content(msg)
None # but the function is typed -> str
```

Fix mirrors the #3375 pattern: `return last_content.text or ""` in `extract_last_content`, `return last_content.text or None` in `extract_last_text` for explicit semantics.

Test plan

• New tests in `tests/utils/test_pretty_print_and_items.py` covering all four cases (None vs string content) for both helpers, mirroring the existing `text_message_output` tests added in fix: guard None text in text_message_output and add output guardrail count to RunErrorDetails #3375.
• `uv run pytest tests/utils/test_pretty_print_and_items.py` — 9 passed.
• `uv run ruff check src/agents/items.py tests/utils/test_pretty_print_and_items.py` — All checks passed.

Issue number

N/A — found while auditing for the same pattern that #3375 fixed.

Checks

• I've added new tests (if relevant)
• I've added/updated the relevant documentation (no doc changes needed)
• I've run `make lint` and `make format`
• I've made sure tests pass

[ioleksiuk]

Thanks for the review! Added inline comments on both extract_last_content (line 687) and extract_last_text (line 701) explaining the rationale — same shape as the existing comment in extract_text (items.py:714-720):

last_content.text is typed as str per the Responses API schema, but provider gateways (e.g. LiteLLM) and model_construct paths during streaming have been observed surfacing None.

The two comments differ slightly in the second sentence to match their respective return types (-> str coerces to "", -> str | None coerces to None so truthiness checks still work).

Force-pushed the amended commit.

[ioleksiuk]

Follow-up: on second look at the diff, the original extract_last_text change (return last_content.text or None) had a subtle semantic regression I missed — it conflated text="" with text=None. Both became None, which would silently change behavior for tools that legitimately return empty text (callers using result is None to distinguish "no content" from "empty content" would now see None for both).

Reverted that change and dropped the related tests/inline comment. The PR is now scoped to a single, real fix:

• extract_last_content (declared -> str) — coerce None to "" so the function never violates its type contract.

extract_last_text left untouched because:

• It's declared -> str | None, so returning None is legal per the signature.
• The pre-existing behavior of returning "" for empty text is meaningful (callers can still distinguish via is None).

Updated commit message + PR title accordingly. Sorry for the noise.

[seratch]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 👍

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".