fix(transcript): record the codex session model from turn_context

[Yurii214]

parse_codex_transcript populated session.model from nowhere — it only handled session_meta, which carries cwd, branch and timestamps but not the model. codex records the turn's model on a separate turn_context record, so the normalized transcript's session.model was always None for codex, while parse_claude_transcript fills it from the assistant message.

this reads the model off turn_context (first non-empty wins, since the model can change across turns), so both agents expose the session model symmetrically. no other field is touched.

verified: pytest tests/, mypy src, ruff check src tests all green. the codex transcript test now asserts the model is extracted.

Summary by CodeRabbit

• Bug Fixes
  • Improved Codex transcript parsing to accurately identify and record the model used during a session.
  • Model information is now captured from turn-level context when available, improving session attribution.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: c7397406-a40b-4bde-9a7a-36c537e3811c

📥 Commits

Reviewing files that changed from the base of the PR and between bfb6b0a and d8e954f.

📒 Files selected for processing (2)

• src/vouch/transcript.py
• tests/test_session_transcript.py

---

Walkthrough

Codex transcript normalization now reads the session model from the first non-empty turn_context payload when unset. The Codex fixture and parser test verify attribution as gpt-5-codex.

Changes

Codex model attribution

Layer / File(s)	Summary
Parse and validate turn-context models src/vouch/transcript.py, tests/test_session_transcript.py	parse_codex_transcript extracts an unset session model from non-empty turn_context data, and the test fixture and assertion validate gpt-5-codex.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested reviewers: plind-junior

🚥 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 clearly describes the main change: recording the Codex session model from turn_context.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.

✨ 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.}

[Yurii214]

Hi @plind-junior — friendly heads-up on this one: it's fully green (tests pass on py3.11/3.12/3.13 and CodeRabbit approved the change), but it can't auto-merge because the coderabbit-gate job 403s when it tries to post the coderabbit-approved status from a fork PR — the exact problem tracked in #511 — so the required check stays red even though the review passed.

Since posting that status / merging past it needs owner permissions, would you be able to merge this manually when you get a chance? It's a small, self-contained fix (codex transcripts drop session.model because the model is recorded on turn_context, not session_meta, so the claude parser fills it and the codex one didn't). Happy to rebase onto the latest test first if that's easier. Thanks!