Conversation
end 工具的参数在 tool_calls 循环中解析一次后未保存， 在后续专门处理段又重新解析，导致同一条修复警告输出两次。 现在循环中保存已解析的参数，后续直接复用。 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
模型 API 调用失败（rate limit、超时、5xx 等）后，请求自动放回 队列最前端优先重试，避免用户消息静默丢失。通过 any_tool_executed 标志确保仅在无副作用时重试，防止消息重复发送。 - 新增 ai_request_max_retries 配置项（默认2，范围0-5） - ModelQueue 新增 retry_queue，调度循环优先处理 - ask() 在首次模型调用失败时向上传播异常 - 热重载注册为需重启项 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
思维链工具调用兼容功能不限于 DeepSeek，GLM 等模型同样适用。 移除所有 DeepSeek 品牌硬编码，修复思维链内容重复日志问题。 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
将是否发送 budget_tokens 从 compat 中独立出来，用户可单独控制是否 在请求中传递思维预算，避免 compat 开启时预算被静默丢弃。 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
移除 send_message 工具的 at_user 参数，改为在消息文本中使用 [@{qq_id}]
格式自动检测并转换为 CQ:at。支持 \[...\] 转义避免触发 @。
转换统一在 MessageSender.send_group_message 中完成。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- 提示词去掉 {QQ号} 占位符写法，用具体示例避免歧义
- 正则兼容 [@{xxx}] 花括号格式作为兜底
- 传入 @ 消息段有昵称时显示为 [@qq(昵称)]
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
支持精确匹配和模糊匹配（子串、拼音首字母/全拼、编辑距离容错）， 按匹配优先级排序返回结果。新增 pypinyin 依赖。 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
关键词也需要转拼音才能与名字拼音做比较，否则中文关键词 无法通过拼音匹配同音/近音的名字（如洛兰匹配紫罗兰）。
save() 是 async 方法，缺少 await 导致协程未执行，end 摘要不会写入磁盘。 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c56f130b4d
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".
src/Undefined/config/loader.py
Outdated
There was a problem hiding this comment.
This parser now only reads thinking_include_budget/thinking_tool_call_compat, so existing deployments that still use deepseek_new_cot_support (or its old env vars) silently fall back to defaults (thinking_include_budget=True, thinking_tool_call_compat=False). For DeepSeek users who previously enabled the old flag to avoid strict thinking payload incompatibilities, this can reintroduce 400 failures immediately after upgrade unless they manually migrate every config key.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Casting count with int(args.get("count", 10) or 10) can raise ValueError when the model emits a non-numeric string (which still happens in function-call arguments despite schema hints), and this occurs before the surrounding try block, so the tool crashes instead of returning a clear validation error. This makes group.find_member brittle on malformed tool arguments.
Useful? React with 👍 / 👎.
1 participant
Summary
end_summary_storage.save()缺少await导致的摘要持久化失败。group.find_member工具，支持按昵称/拼音检索群成员，完善群工具能力与相关配置。