fix: validate list-valued agent outputs

[jbeckwith-oai]

This pull request fixes list-valued agent outputs being treated as conversation history instead of generated content. Output checks now receive the text of every top-level list item, preserving literal characters within string items and leaving the original returned output unchanged.

Input conversation extraction and non-list output handling remain unchanged. Regression coverage exercises the real Agents SDK runner with mocked provider responses under both strict and default error handling.

Validation: two independent reviews; formatting, Ruff, mypy, and pyright passed; 1,596 pytest tests and four release-workflow tests passed.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-24T12:25:24.140568Z	6286fd0	New commits
🔒 Security Review	✅ Completed 2026-09-24T12:27:19.330077Z	6286fd0	New commits

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[markstuart-oai]

Reviewed 213da72 against the released Agents output contract and the existing guardrail adapter. No actionable correctness or structural findings: generated lists now pass all top-level items through the existing text-check policy, while input history, non-list handling, literal string contents, and returned output values remain intact. The focused real-SDK regression coverage exercises both blocked output and unchanged safe results.

Source-only review; all eight checks on this head passed (Python 3.11–3.14 and CodeQL). I did not rerun tests locally.