[delete-cloud]

o6n 实测发现并修复:

• codex provider 把 pipeline 传入的 thinking_config 原样塞进请求体,后端 400 Unsupported parameter → 翻译为 Responses 的 reasoning.effort,disabled 时省略
• gpt-5.2 已被上游下线(ChatGPT 账号现只接受 gpt-5.5/gpt-5.4)→ CONTEXT_SIZES 和 webui 预设同步移除

测试:test_codex_responses 6 个全过(新增 2 个:翻译/省略),vitest 100 全过,ruff 干净

Summary by CodeRabbit

• New Features

  • Improved reasoning controls for Codex requests, translating enabled thinking settings into the appropriate reasoning effort.

• Updates

  • Removed the unavailable GPT-5.2 model option from Codex model selection.
  • GPT-5.5 and GPT-5.4 remain available.

• Bug Fixes

  • Requests now omit reasoning settings when thinking is disabled.

[coderabbitai]

📝 Walkthrough

Walkthrough

The Codex provider now translates enabled thinking_config into Responses API reasoning settings and removes the internal field from requests. The gpt-5.2 model is removed from provider metadata and web UI presets.

Changes

Codex provider and model updates

Layer / File(s)	Summary
Map thinking configuration to reasoning fields coding-agent/src/coding_agent/providers/codex_responses.py, coding-agent/tests/providers/test_codex_responses.py	_build_request removes thinking_config and maps enabled settings to reasoning.effort. Tests cover enabled and disabled configurations.
Remove gpt-5.2 from model availability coding-agent/src/coding_agent/providers/codex_responses.py, coding-agent/webui/app/src/components/Header.tsx, coding-agent/webui/app/src/components/Header.test.tsx	The provider context-size mapping and Codex model presets no longer include gpt-5.2.

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

Possibly related PRs

• delete-cloud/Coding-Agent-Loop#415: Directly modifies codex_responses.py and its tests for thinking_config and reasoning.effort handling.

🚥 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 summarizes the Codex thinking_config translation and model list updates.
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 docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/codex-thinking-config

---

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.}

[coderabbitai]

🧹 Nitpick comments (1)

coding-agent/tests/providers/test_codex_responses.py (1)

129-156: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Cover the default reasoning effort.

The provider defaults enabled thinking to "medium" when effort is absent. This test only supplies "high". Add a case with thinking_config={"enabled": True} and assert body["reasoning"] == {"effort": "medium"}.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@coding-agent/tests/providers/test_codex_responses.py` around lines 129 - 156,
Extend test_stream_translates_thinking_config_to_reasoning_effort with a case
using thinking_config={"enabled": True} and no effort value, then assert the
captured request body contains reasoning set to {"effort": "medium"} while
excluding thinking_config.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@coding-agent/tests/providers/test_codex_responses.py`:
- Around line 129-156: Extend
test_stream_translates_thinking_config_to_reasoning_effort with a case using
thinking_config={"enabled": True} and no effort value, then assert the captured
request body contains reasoning set to {"effort": "medium"} while excluding
thinking_config.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b21db767-58d2-44f7-b554-a7227ec77c3f

📥 Commits

Reviewing files that changed from the base of the PR and between 34c0dc5 and 172ff80.

📒 Files selected for processing (4)

• coding-agent/src/coding_agent/providers/codex_responses.py
• coding-agent/tests/providers/test_codex_responses.py
• coding-agent/webui/app/src/components/Header.test.tsx
• coding-agent/webui/app/src/components/Header.tsx