ChatSpeed v2.0.9

[2.0.9]

🚀 New Features

• More controlled Workflow tools and context:
  • MCP servers can now expose selected tools to Workflow individually, with authorization requested only when a tool actually needs access.
  • Workflow supports file references in prompts and offers clearer choices when an agent needs to create files.
  • Added a Skills entry to the Workflow input menu for quicker access to installed capabilities.
  • Completion reports can now include an explicit summary, and todo snapshots from planning remain separate from execution todos.
• Workflow navigation and tool results:
  • Added a compact sidebar rail to give Workflow conversations more usable space while keeping navigation accessible.
  • Large tool output is preserved safely, while Shell output is reduced more intelligently for Git, tests, builds, Rust/Go, JSON, logs, and compound commands.
• Model API-key protection:
  • Model API keys are protected in user key files and can be unlocked when required, strengthening local credential protection.

🪄 Improvements

• Workflow reliability and clarity:
  • Improved completion-report rendering, tool status localization, approval details, and the display of tools waiting for execution.
  • Refined Workflow recovery, queue and backup restoration, message navigation, task-window folding, continuation deletion, and terminal/manual-clear state reconciliation.
  • Hardened approved-plan transitions, approval recovery, structured frontend tool states, inherited agent configuration, and concurrent sub-agent execution.
• CCProxy protocol compatibility and diagnostics:
  • Improved OpenAI Responses API handling, including request and response conversion across supported backends.
  • Proxy errors are now converted into the correct response shape for OpenAI, Claude, Gemini, and Ollama clients, with clearer backend transport error reporting.

🐞 Bug Fixes

• Workflow and tool-output correctness:
  • Fixed incomplete preservation of tool output in Workflow context and restored correct output display after Shell-result reduction.
  • Fixed sidebar file-path references, approval UI contracts, runtime observation recovery, and Workflow execution/replay edge cases.
• Workflow data integrity:
  • Prevented a potential store write-lock deadlock when sub-agents run concurrently and improved cleanup of deleted Workflows.

---

[2.0.9]

🚀 新功能

• 更可控的 Workflow 工具与上下文：
  • MCP 服务器现在可逐个选择向 Workflow 暴露的工具，并且仅在工具实际需要访问权限时请求授权。
  • Workflow 支持在提示词中引用文件；当 Agent 需要创建文件时，提供更清晰的创建方式选择。
  • Workflow 输入菜单新增 Skills 入口，可更快捷地使用已安装的能力。
  • 完成报告现在可包含明确摘要；计划阶段的待办快照会与执行阶段的待办保持分离。
• Workflow 导航与工具结果：
  • 新增紧凑侧边栏轨道，在保留导航入口的同时为 Workflow 对话提供更多可用空间。
  • 大型工具输出会被安全保留；同时针对 Git、测试、构建、Rust/Go、JSON、日志和复合命令，更智能地精简 Shell 输出。
• 模型 API Key 保护：
  • 模型 API Key 现由用户密钥文件保护，并可在需要时解锁，进一步增强本地凭据安全性。

🪄 改进

• Workflow 可靠性与清晰度：
  • 改进完成报告渲染、工具状态本地化、审批详情，以及等待执行工具的展示。
  • 优化 Workflow 恢复、队列与备份还原、消息导航、任务窗口折叠、删除续接后的状态回退，以及结束状态与手动清空状态的协调。
  • 加固已批准计划的过渡、审批恢复、前端结构化工具状态、继承的 Agent 配置和并发子 Agent 执行。
• CCProxy 协议兼容性与诊断：
  • 改进 OpenAI Responses API 的处理，包括在已支持后端之间进行请求与响应转换。
  • CCProxy 现在会针对 OpenAI、Claude、Gemini 和 Ollama 客户端转换为正确的错误响应格式，并提供更清晰的后端传输错误信息。

🐞 修复

• Workflow 与工具输出正确性：
  • 修复 Workflow 上下文中工具输出未被完整保留的问题，并恢复 Shell 结果精简后的正确展示。
  • 修复侧边栏文件路径引用、审批 UI 契约、运行时观察结果恢复，以及 Workflow 执行与回放中的边界问题。
• Workflow 数据完整性：
  • 防止子 Agent 并发执行时可能发生的存储写锁死锁，并优化删除 Workflow 时的清理流程。