Commit Graph

266 Commits

Author SHA1 Message Date
jxxghp
140d224a9a fix agent stream blocking during command execution
Offload synchronous message edits from the event loop and stream subprocess output so long-running commands stay responsive.
2026-04-27 07:57:32 +08:00
jxxghp
bae086d8b8 更新 __init__.py 2026-04-27 06:57:18 +08:00
jxxghp
221eb21694 refine internal middleware llm usage for streaming agents
Use a non-streaming model for middleware-only calls so internal outputs do not leak into user streams and model-based middleware stays consistent.
2026-04-27 06:55:41 +08:00
jxxghp
4208c79d72 refine tool提示语为更简洁风格,补充last_buffer_char属性及非VERBOSE模式流式输出换行逻辑,新增工具流式分隔符单元测试 2026-04-26 11:15:11 +08:00
jxxghp
90245a13e1 refine non-verbose prompt wording 2026-04-26 08:54:07 +08:00
jxxghp
b5979b9b09 refine agent subscription defaults and silent tool prompts 2026-04-26 08:51:56 +08:00
jxxghp
0277288a41 feat: add agent session usage status reporting
Track per-session model and token usage so users can inspect context pressure and cumulative usage with /session_status.
2026-04-26 08:19:05 +08:00
jxxghp
79bfeaf2af 移除工具调用前的流重置,保留模型思考文本可见 2026-04-25 23:12:34 +08:00
jxxghp
4fe41ba5e9 更新 base.py 2026-04-25 22:16:15 +08:00
jxxghp
14d6e2febc Refine agent prompts for concise professional replies 2026-04-25 22:04:35 +08:00
jxxghp
97c7e71207 更新 Agent Prompt.txt 2026-04-25 21:51:47 +08:00
jxxghp
bfc27d151c 更新 ask_user_choice.py 2026-04-25 11:36:36 +08:00
jxxghp
f2b56b8f40 更新 ask_user_choice.py 2026-04-25 11:35:32 +08:00
jxxghp
cefb60ba2c refactor: unify message interactions 2026-04-22 15:18:04 +08:00
jxxghp
8c256d91bd refine custom identifier skill scope 2026-04-21 17:31:37 +08:00
jxxghp
ae15eac0f8 feat: normalize internal system user ID in notification dispatch
- Add SYSTEM_INTERNAL_USER_ID constant and helpers to app.utils.identity
- Ensure internal user ID is normalized to None before dispatching notifications, preventing misrouting to external channels
- Refactor MessageChain to use normalization for all message dispatch methods
- Add tests for internal user ID normalization and notification dispatch behavior
2026-04-21 14:32:14 +08:00
jxxghp
e0e21e39a2 refactor: generalize agent interaction requests 2026-04-16 22:51:51 +08:00
jxxghp
cc31c66b93 feat: add agent button choice workflow 2026-04-16 22:32:59 +08:00
jxxghp
04c2a1eb18 Add manual AI redo flow 2026-04-15 17:10:18 +08:00
jxxghp
13c3c082b8 Improve agent image capability routing 2026-04-15 08:55:32 +08:00
jxxghp
7a5e513f25 feat(agent): support file attachments and local file replies 2026-04-14 15:22:01 +08:00
jxxghp
e5f97cd299 feat(agent): add voice message support with TTS/STT for Telegram and WeChat
- Integrate voice message handling: detect and extract audio references from Telegram and WeChat messages, route to agent with voice reply preference.
- Add voice provider abstraction and OpenAI-based TTS/STT implementation.
- Implement agent tool `send_voice_message` for generating and sending voice replies, with fallback to text if voice is unavailable.
- Extend agent prompt and context to support voice reply instructions.
- Update notification and message schemas to support audio fields.
- Add Telegram and WeChat voice sending logic, including audio file conversion and temporary media upload for WeChat.
- Add tests for voice helper and agent voice routing.
2026-04-12 12:30:02 +08:00
jxxghp
6c5fae56d9 Add agent image support for Telegram and Slack 2026-04-11 20:40:02 +08:00
jxxghp
bf12a8679d refactor: 移除 agent 批量重试逻辑中的多余 try 块并优化缩进 2026-04-10 15:03:49 +08:00
jxxghp
75fca971d4 refactor(agent): 重命名 can_edit_message 为 is_auto_flushing 更贴切语义 2026-04-09 23:29:29 +08:00
jxxghp
22f3244bf5 fix(agent): 流式+啰嗦模式下渠道不支持编辑时立即发送工具消息
渠道不支持编辑时没有定时刷新任务,emit 到 buffer 的内容不会被推送。
新增 can_edit_message 属性区分两种模式:支持编辑的继续 emit 到 buffer,
不支持编辑的 take 出 agent 文字与工具消息合并独立发送。
2026-04-09 23:26:39 +08:00
jxxghp
aafc4b3a39 fix(agent): start_streaming 始终标记流式状态以支持 buffer 收集
渠道不支持消息编辑时,仍需标记 streaming_enabled 为 True,
以便啰嗦模式下工具调用时能通过 is_streaming 进入流式分支
发送 agent 中间文字和工具消息。只是不启动定时刷新任务。
2026-04-09 23:19:34 +08:00
jxxghp
18906e5ab2 更新 __init__.py 2026-04-09 22:51:37 +08:00
jxxghp
9675d199f9 fix(agent): 非流式模式下不发送任何工具中间消息 2026-04-09 22:48:03 +08:00
jxxghp
78e8faa203 fix(agent): 非流式模式下啰嗦模式仍需发送工具调用中间消息
啰嗦模式+渠道不支持编辑时,虽然 is_streaming 为 False,
但 astream 仍会将 token 写入 buffer,需要在工具调用时
取出 agent 文字与工具消息合并发送
2026-04-09 22:23:00 +08:00
jxxghp
d5ed9bc654 fix(agent): 简化非流式模式下工具调用的消息处理逻辑
非流式模式下使用 ainvoke 执行,无流式 token 产出,
不需要操作 stream_handler 或发送中间消息
2026-04-09 22:20:09 +08:00
jxxghp
770065d9ed feat(agent): 优化Agent流式输出与工具消息发送逻辑
- 新增 _should_stream() 方法,根据运行环境决定是否启用流式输出:
  后台模式不启用;渠道支持编辑启用;啰嗦模式开启时也启用
- 非流式模式下使用非流式LLM + ainvoke,避免不必要的流式开销
- 非啰嗦模式下工具调用时不发送任何中间消息(agent文字和工具提示),直接清掉缓冲区
2026-04-09 22:12:20 +08:00
jxxghp
dc428e7de0 feat(skills): 内置技能支持版本号管理,更新时自动覆盖旧版本
SKILL.md frontmatter新增version字段,同步时比较版本号,
内置版本更高时直接覆盖用户目录中的旧版本。
2026-04-09 07:17:04 +08:00
jxxghp
0c51d79be7 feat(agent): 合并同批次整理失败的agent重试调用,避免重复浪费token
同一download_hash或同一源目录下的失败记录在5分钟缓冲期内合并为一次agent调用,
批量处理时只识别一次媒体信息后复用到所有文件。
2026-04-09 07:16:56 +08:00
jxxghp
3c7cd2186f 查询订阅历史工具有名称过滤时不分页直接返回所有匹配结果 2026-04-08 07:58:54 +08:00
jxxghp
5acfd683b9 agent工具支持翻页及取消数量限制 2026-04-08 07:41:34 +08:00
jxxghp
6b01901a4a 更新 search_web.py 2026-04-08 07:29:30 +08:00
jxxghp
1ca54afd6c 更新 search_person.py 2026-04-08 07:27:29 +08:00
jxxghp
9c75c2d22e 更新 search_media.py 2026-04-08 07:26:54 +08:00
jxxghp
79ec3ed2c3 更新 list_directory.py 2026-04-08 07:21:37 +08:00
jxxghp
7072d2cfe8 更新 query_installed_plugins.py 2026-04-08 07:15:13 +08:00
jxxghp
c0c08b0b84 更新 query_subscribe_history.py 2026-04-08 07:12:39 +08:00
jxxghp
01329195ee 更新 query_subscribes.py 2026-04-08 07:11:45 +08:00
jxxghp
1e338e48ab fix(agent): 基于langgraph_step过滤中间步骤思考文本,抽离ThinkTagStripper类
- 利用metadata中的langgraph_step检测工具调用前的中间步骤,非VERBOSE模式下
  自动reset清除模型输出的计划/推理文本(如NEXT STEPS、tool call描述等)
- 将<think>标签流式剥离逻辑抽离为独立的_ThinkTagStripper类,简化主流程
2026-04-07 12:42:46 +08:00
jxxghp
ac9c9598f4 feat(agent): add tools for querying and updating custom identifiers 2026-04-07 09:00:15 +08:00
jxxghp
8109ffb445 feat(agent): add /stop_agent command for emergency stop of agent reasoning
Add /stop_agent command that cancels the currently running agent reasoning
task without clearing the session or memory. Unlike /clear_session which
destroys the entire session, this allows users to stop a long-running or
stuck agent process and continue the conversation afterward.
2026-04-07 07:32:35 +08:00
jxxghp
902394f86e fix(agent): resolve circular import by lazy-importing Command in run_slash_command and list_slash_commands 2026-04-07 00:16:09 +08:00
jxxghp
9fefd807f9 refactor(agent): rename list_all_commands to list_slash_commands and skill to command-dispatch 2026-04-07 00:00:10 +08:00
jxxghp
a8fb4a6d84 refactor(agent): rename run_plugin_command to run_slash_command to avoid confusion with execute_command (shell) 2026-04-06 23:53:49 +08:00
jxxghp
7806267e92 feat(agent): add command-execute skill for intelligent command dispatch
- Enhance run_plugin_command tool to support all registered commands (system preset + plugin + other), not just plugin commands
- Add list_all_commands tool to discover all available commands with descriptions and categories
- Add command-execute skill that guides the agent to recognize user intent from natural language and match it to available system/plugin commands
2026-04-06 23:45:48 +08:00