fix(tools): 修复工具系统的多个问题

[69gg]

修复内容

1. 修复工具名误加前缀问题

• 问题：skills/registry.py 在注册工具时使用了原始 config 而不是处理后的 item.config
• 影响：导致工具名被重复添加前缀（如 memory.memory.add）
• 修复：使用 item.config 替代原始 config，确保工具名正确

2. 修复工具配置中的分隔符问题

• 问题：toolsets 的 config.json 中工具名包含了类别前缀（如 memory.add）
• 影响：与注册逻辑冲突，导致工具名格式错误
• 修复：移除 config.json 中的前缀，由注册系统自动添加
• 涉及工具：
  • memory: add, delete, list, update
  • notices: get, list, stats
  • messages: send_group_sign

3. 完善上下文注入

• 新增：向工具 context 注入 memory_storage 和 runtime_config
• 位置：ai_coordinator.py 和 scheduler.py
• 目的：使工具能够访问记忆存储和运行时配置

4. 优化 naga agent 文件读取工具

• 新增参数：max_chars - 可选的最大字符数限制
• 改进截断逻辑：
  • 从硬编码的 10000 字符改为可配置限制
  • 提供更清晰的截断提示信息（显示总字符数和截断位置）
• 建议用法：小文件不指定，大文件可设置为 50000-100000

5. 修复 end 工具的消息发送检查

• 新增参数：force - 强制结束对话，跳过未发送消息检查
• 新增检查：如果有 summary 但本轮未发送消息，拒绝执行（除非 force=true）
• 新增标记：在所有消息发送处设置 context["message_sent_this_turn"] = True
• 涉及文件：
  • tools/end/handler.py - 添加检查逻辑
  • toolsets/messages/send_message/handler.py - 添加标记
  • toolsets/messages/send_private_message/handler.py - 添加标记

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1d251522cd

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".

[chatgpt-codex-connector]

Account for non-text send tools before blocking end

This new guard blocks end whenever message_sent_this_turn is false, but in this commit only send_message/send_private_message set that flag. Other user-facing tools (for example render_markdown and render_latex) can already send content via sender/send_image_callback without updating this key, so a valid flow of "send image, then end with summary" is now rejected and can force extra model loops or force=true workarounds.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Validate force as bool before bypassing end guard

The bypass uses force = args.get("force", False) and then checks if summary and not force, so any truthy non-boolean value (e.g. a model-emitted string "false") disables the protection. Because parse_tool_arguments preserves value types, malformed but common tool-call payloads can skip the unsent-message check unintentionally.

Useful? React with 👍 / 👎.