Commit Graph

109 Commits

Author SHA1 Message Date
LIghtJUNction
26c5e67efe Refactor: change relative imports to absolute imports in tool_loop_agent_runner.py 2026-03-21 14:13:43 +08:00
LIghtJUNction
cefeaf8d9f Fix: add task tracking to BaseAgentRunner and ToolLoopAgentRunner, replace asdict with custom serialization in Response 2026-03-21 13:55:05 +08:00
LIghtJUNction
61e525afd4 Merge branch 'master' into dev 2026-03-21 12:25:04 +08:00
LIghtJUNction
b2a04ffed1 Fix syntax error in tool_loop_agent_runner.py 2026-03-21 04:21:21 +08:00
LIghtJUNction
3405c72b5e Modify context configuration parameters
Updated max_context_tokens default value and adjusted enforce_max_turns logic.
2026-03-21 04:17:59 +08:00
LIghtJUNction
4ee93c7f83 Limit max_step to a maximum of 3
基本都是设置为30,30太夸张了,这是日志里面截出来的:
system,user,assistant,user,assistant,user,assistant,tool,assistant,user,assistant,user,assistant,user,assistant,tool,assistant,user,assistant,user,assistant,user,assistant,user,assistant,user,assistant,user,assistant,user,assistant,tool,assistant,user,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,user,assistant,user,assistant,user,assistant,user,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,assistant,user,assistant,tool,assistant,user,assistant,user,assistant,user,assistant,user,assistant,tool,assistant,user,assistant,tool,assistant,user,assistant,tool,assistant,user,assistant,user,assistant,user,assistant,user,assistant,user,assistant,tool,assistant,user,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,tool,assistant,user,assistant,tool,tool,assistant,tool,assistant,user,assistant,tool,assistant,user,assistant,user,assistant,user,assistant,user,assistant,user,assistant,user
2026-03-21 04:12:49 +08:00
LIghtJUNction
1b92c517e7 Merge branch 'master' into dev 2026-03-21 00:31:18 +08:00
晴空
d2e0bc778a fix: preserve PATHEXT for stdio mcp servers on windows (#5822)
* fix: preserve PATHEXT for stdio mcp servers on windows

* chore: delete test_mcp_client.py

---------

Co-authored-by: Soulter <905617992@qq.com>
2026-03-21 00:27:54 +08:00
whatevertogo
2c279abad1 fix: prevent accidental removal of MCP external tools due to name collisions with disabled built-in tools (#5925)
* fix: 解决 MCP 工具与内置工具重名时的连坐问题

- 修改 get_func 方法:优先返回已激活的工具
- 修改 get_full_tool_set 方法:使用 add_tool 防止同名冲突
- 修改 add_tool 方法:优先保留已激活的工具

Fixes #5821

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor: address PR review feedback for tool conflict resolution

- Fix inconsistency: get_func now uses reversed() to match ToolSet.add_tool's
  "last-active-wins" logic, preventing potential "tool hijacking" issues
- Improve readability: replace double negative condition with clearer logic
- Add compatibility: use getattr with default for tools without 'active' attribute
- Remove unnecessary deepcopy: MCPTool runtime objects should not be deep copied
- Update docstring: accurately describe the actual tool resolution behavior

Addresses review comments from sourcery-ai, gemini-code-assist, and Copilot.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* test: add tests for tool conflict resolution (issue #5821)

Add comprehensive tests for ToolSet.add_tool, get_func, and get_full_tool_set
to verify the conflict resolution behavior when MCP tools share names with
built-in tools.

Test cases:
- ToolSet.add_tool: active/inactive priority, last-one-wins for same state
- get_func: returns last active tool, fallback to last matching tool
- get_full_tool_set: deduplication logic, no deepcopy, MCP overrides disabled builtin

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: 修复工具冲突处理逻辑,确保未激活工具不被错误移除

---------

Co-authored-by: whatevertogo <whatevertogo@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-20 23:44:07 +08:00
LIghtJUNction
4f04d39348 同步:同步主线 2026-03-20 16:59:43 +08:00
leonforcode
0b69034491 fix: prevent truncation logic from removing the only user message in long tool-calling conversations (#6198)
* fix: 压缩算法删除 user 消息 Bug 修复

* perf: improve truncate algo

---------

Co-authored-by: Soulter <905617992@qq.com>
2026-03-20 12:05:54 +08:00
Yufeng He
e286da75c4 fix: 截断器丢失唯一 user 消息导致智谱等 provider 返回 400 (#6581)
* fix: 截断器丢失唯一 user 消息导致 API 400

修复 #6196

当对话只有一条 user 消息(长 tool chain 场景:system → user → assistant
→ tool → assistant → tool → ...),三个截断方法都会把这条 user 消息丢掉,
导致智谱、Gemini 等要求 user 消息的 provider 返回 400。

改动:
- 提取 `_split_system_rest()` 去掉三个方法里重复的 system/non-system 拆分
- 新增 `_ensure_user_message()`:截断后如果没有 user 了,从原始消息里补回
  第一条 user,避免违反 API 格式要求
- 删掉 `truncate_by_dropping_oldest_turns` 里把没有 user 就清空全部消息的逻辑
- 5 个新测试覆盖单 user + 长 tool chain 场景,3 个旧测试更新断言

* style: format code

---------

Co-authored-by: Yufeng He <40085740+universeplayer@users.noreply.github.com>
Co-authored-by: RC-CHN <1051989940@qq.com>
2026-03-20 10:06:20 +08:00
LIghtJUNction
92bae1fdae docs: replace prints with logger in upload_doc_images_to_r2.py 2026-03-19 23:30:25 +08:00
LIghtJUNction
a3371ad6c8 core: refactor modules & fix typing/runtime issues 2026-03-19 23:06:27 +08:00
LIghtJUNction
d16f62423c Merge branch 'master' into dev 2026-03-19 19:25:47 +08:00
Yufeng He
2cb6c84eeb feat: context token counting support for multimodal content (images, audio, and chain-of-thought) (#6596)
EstimateTokenCounter 之前只计算 TextPart,完全忽略 ImageURLPart、
AudioURLPart 和 ThinkPart。多模态对话中图片占 500-2000 token,
不被计入会导致 context 压缩触发过晚,API 先报 context_length_exceeded。

改动:
- ImageURLPart 按 765 token 估算(OpenAI vision 低/高分辨率中位数)
- AudioURLPart 按 500 token 估算
- ThinkPart 的文本内容正常计算
- 10 个新测试覆盖各类型单独和混合场景

Co-authored-by: Yufeng He <40085740+universeplayer@users.noreply.github.com>
2026-03-19 17:25:49 +08:00
LIghtJUNction
14ec513b0d fix: clean lint suppressions and async route errors 2026-03-18 18:06:42 +08:00
LIghtJUNction
4bf2f0cb28 fix(dashboard): handle missing static assets gracefully and fix async/ruff issues
- Dashboard: Catch missing index.html error, log warning, and disable WebUI instead of crashing.
- Dashboard: Use anyio.Path for async file existence checks in SSL config.
- CLI/Backup: Replace blocking file operations with anyio async operations.
- Core: Replace blocking file operations with anyio in Coze/Dify clients and TTS simulation.
- Provider: Rename 'timeout' param to 'init_timeout'/'request_timeout' to fix ASYNC109 warnings.
- Ruff: Fix various ASYNC230/ASYNC240 errors across multiple files.
2026-03-18 16:46:22 +08:00
LIghtJUNction
6dc13b880b refactor: rename skills_like to lazy_load and add skill schema support 2026-03-17 21:32:06 +08:00
LIghtJUNction
f3d6f762c3 refactor: rename skills_like to lazy_load and add skill schema support 2026-03-17 20:46:07 +08:00
LIghtJUNction
4b7d42c2a3 chore: ruff format 2026-03-17 16:53:45 +08:00
LIghtJUNction
f6321be8c8 sync 2026-03-17 16:51:45 +08:00
whatevertogo
1fd3d4ce0e fix: subagent lookup failure when using default persona (#5672)
* fix: resolve subagent persona lookup for 'default' and unify resolution logic

- Add PersonaManager.get_persona_v3_by_id() to centralize v3 persona resolution
- Handle 'default' persona_id mapping to DEFAULT_PERSONALITY in subagent orchestrator
- Fix HandoffTool.default_description using agent_name parameter correctly
- Add tests for default persona in subagent config and tool deduplication

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor: simplify get_default_persona_v3 using get_persona_v3_by_id

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: whatevertogo <whatevertogo@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Soulter <37870767+Soulter@users.noreply.github.com>
2026-03-17 15:42:15 +08:00
ccsang
8e6c835b85 refactor(agent): extract image-handling logic into helper function
Address Sourcery AI review feedback: the image-handling logic was
duplicated for ImageContent and EmbeddedResource cases.

Changes:
- Extract _handle_image_content() helper function
- Consolidate image caching, result appending, and yielding logic
- Reduce code duplication and improve maintainability
2026-03-16 00:17:33 +00:00
ccsang
fb2a2a63f2 fix(agent): process all content items in CallToolResult, not just the first
Fixes #6140

When a tool returns CallToolResult with multiple content items (e.g.,
both TextContent and ImageContent), the agent was only processing
content[0], ignoring the rest.

Changes:
- Replace direct content[0] access with enumerate(res.content) loop
- Process all content items: TextContent, ImageContent, EmbeddedResource
- Use content_index for image caching to distinguish multiple images

This fixes the issue where tools like Bilibili plugin return both
text descriptions and screenshots, but LLM only received one of them.
2026-03-16 00:17:33 +00:00
zenfun
e85eef05b8 fix: stabilize tool injection for LLM prefix cache hits
Two changes to make the tool schema sent to the LLM deterministic:
1. ToolSet.normalize() — sort tools by name before serialization.
   Called at the end of build_main_agent() after all injection passes.
   Eliminates ordering drift from plugin load order, MCP reconnection,
   and persona tool list differences.
2. Always inject full sandbox tool set — ComputerToolProvider now
   returns get_default_sandbox_tools() unconditionally, regardless of
   sandbox boot state. Browser tools are always in the schema even if
   the sandbox profile lacks browser capability. The executor rejects
   calls to unavailable browser tools with a descriptive error instead
   of silently omitting them from the schema.
   This eliminates the pre-boot/post-boot tool set jump that caused
   prefix cache misses on the second request of a conversation.
2026-03-12 02:43:19 +08:00
Gao Jinzhe
42b8293f99 Merge branch 'AstrBotDevs:master' into agent-fix-clean 2026-03-11 09:48:36 +08:00
Soulter
3379587223 feat(mcp): enhance logging and initialize MCP clients in background (#5993)
* feat(mcp): enhance logging and initialize MCP clients in background

fixes: #5777

* rf

* fix(mcp): simplify MCP client initialization in background

* fix(mcp): update error message for MCP background initialization failure
2026-03-11 00:00:48 +08:00
邹永赫
e25a1a42cf Revert "fix: clarify missing MCP stdio command errors (#5992)"
This reverts commit 0c771e4a77.
2026-03-11 00:08:06 +09:00
エイカク
0c771e4a77 fix: clarify missing MCP stdio command errors (#5992)
* fix: clarify missing MCP stdio command errors

* refactor: tighten MCP error presentation helpers

* fix: improve MCP test connection feedback

* fix: structure MCP test connection errors

* refactor: share MCP test error codes
2026-03-10 23:05:50 +09:00
advent259141
ff4412a627 refactor: Centralize and decouple computer-use tool injection logic into a new ComputerToolProvider and associated tool modules. 2026-03-10 22:00:23 +08:00
whatevertogo
3fd6c4c8a6 fix: 修复 asyncio 事件循环相关问题 (#5774)
* fix: 修复 asyncio 事件循环相关的问题

1. components.py: 修复异常处理结构错误
   - 将 except Exception 移到正确的内部 try 块
   - 确保 _download_file() 异常能被正确捕获和记录

2. session_lock.py: 修复跨事件循环 Lock 绑定问题
   - 添加 _access_lock_loop_id 追踪事件循环
   - 当事件循环变化时重新创建 Lock

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: 根据代码审查反馈修复问题

1. components.py: 移除 asyncio.set_event_loop() 调用
   - 创建临时 event loop 时不再设置为全局
   - 避免干扰其他 asyncio 使用

2. session_lock.py: 简化延迟初始化逻辑
   - 移除 loop-ID 追踪和 _get_lock 方法
   - 使用 setdefault 简化 session lock 创建
   - 保留延迟初始化行为

3. wecomai_queue_mgr.py: 使用 time.monotonic() 替代 loop.time()
   - 同步方法不再依赖活动的 event loop
   - 避免在非异步上下文中抛出 RuntimeError

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: 优化 asyncio 事件循环管理,使用安全的方式创建和关闭事件循环

* fix: 根据代码审查反馈改进异常处理和事件循环使用

- main.py: 显式处理 check_dashboard_files() 返回 None 的情况
- components.py: 使用 logger.exception 保留异常堆栈信息
- star_manager.py: 添加 Future 异常回调处理 __del__ 执行异常
- bay_manager.py: 缓存事件循环引用避免重复调用

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor: 简化 SessionLockManager 使用 defaultdict 和 setdefault

- 使用 defaultdict(asyncio.Lock) 简化锁的懒创建
- 使用 setdefault 简化 _get_loop_state 逻辑
- 减少 get + if 分支,提升可读性

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: 降低 webui_dir 检查失败时的日志级别为 warning

改为警告而非退出,允许程序在无 WebUI 的情况下继续运行

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor: 重构事件循环锁管理,简化锁状态管理逻辑

* 新增对 SessionLockManager 的多事件循环隔离测试

* fix: 修复测试中的变量声明和断言,确保事件循环管理器的正确性

* fix: 修复插件删除时异常处理逻辑,确保正确记录错误信息

* fix: 新增针对多个事件循环的 OneBot 实例的测试,确保锁对象在不同事件循环间不共享

---------

Co-authored-by: whatevertogo <whatevertogo@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-09 01:00:13 +09:00
エイカク
451ad685ae feat: 集成 DeerFlow Agent Runner 并优化流式处理 (#5581)
* feat: integrate DeerFlow agent runner and improve stream handling

* refactor: split DeerFlow stream flow and close stale client on reset

* fix: enforce max_step and correct timeout type check

* fix: harden DeerFlow config parsing and session lifecycle

* fix: preserve third-party runner error semantics and harden image parsing

* perf: bound DeerFlow values history and seen-id cache

* refactor: improve deerflow stream semantics and client lifecycle

* fix: harden third-party runner error semantics and fallback aggregation

* refactor: reduce deerflow image log noise and lazy-init api session

* perf: avoid unnecessary iterable copies in deerflow stream utils

* refactor: centralize runner error key and clarify deerflow client lifecycle

* refactor: simplify third-party runner output flow

* fix: defer streaming runner cleanup and unify error mapping

* fix: handle id-less values messages and redact stream payload logs

* fix: improve deerflow error signaling and third-party runner flow

* fix: support deerflow proxy and refine runner lifecycle

* fix: tighten deerflow image validation and runner lifecycle

* feat: support deerflow image output components

* fix: harden runner stream cleanup and refactor deerflow config

* fix: preserve deerflow done hook and simplify runner lifecycle

* refactor: simplify third-party runner aggregation and lifecycle closing

* fix: preserve first deerflow values payload and simplify runner flow

* refactor: unify runner final resolution and harden deerflow close state

* refactor: share int coercion and make deerflow close best effort

* refactor: extract deerflow mappers and streamline third-party lifecycle

* refactor: simplify third-party flow and harden sse flush parsing

* fix: make deerflow runner close path best effort

* refactor: simplify third-party orchestration and centralize deerflow keys

* refactor: simplify third-party chunk flow and deerflow finalization

* fix: harden deerflow stream parsing and simplify runner lifecycle

* refactor: remove redundant deerflow values text assignment

* fix: improve deerflow timeout diagnostics and image input feedback

* refactor: flatten third-party runner lifecycle and aggregation

* chore: use deerflow official remote svg icon

* chore: remove unused deerflow local logo asset
2026-03-01 12:31:38 +09:00
エイカク
7bf44bd8d2 feat: support persona custom error reply message with fallback (#5547)
* feat: support persona custom error reply message with fallback

* refactor: centralize persona custom error message helpers
2026-02-28 13:34:12 +09:00
Soulter
5219ba5c4e feat: implement follow-up message handling in ToolLoopAgentRunner (#5484)
* feat: implement follow-up message handling in ToolLoopAgentRunner

* fix: correct import path for follow-up module in InternalAgentSubStage
2026-02-26 21:38:47 +08:00
Luna_Dol
6d76d55452 fix: ensure tool call/response pairing in context truncation (#5417)
* fix: ensure tool call/response pairing in context truncation

* refactor: simplify fix_messages to single-pass state machine
2026-02-25 15:21:30 +08:00
Chen
997b51102b feat: add image urls / paths supports for subagent (#5348)
* fix: 修复5081号PR在子代理执行后台任务时,未正确使用系统配置的流式/非流请求的问题(#5081)

* feat:为子代理增加远程图片URL参数支持

* fix: update description for image_urls parameter in HandoffTool to clarify usage in multimodal tasks

* ruff format

---------

Co-authored-by: Soulter <905617992@qq.com>
2026-02-23 22:16:14 +08:00
Soulter
e357d9de74 feat: add stop functionality for active agent sessions and improve handling of stop requests (#5380)
* feat: add stop functionality for active agent sessions and improve handling of stop requests

* feat: update stop button icon and tooltip in ChatInput component

* fix: correct indentation in tool call handling within ChatRoute class
2026-02-23 20:21:30 +08:00
Gao Jinzhe
8a8ec492d7 feat: supports spawn subagent as a background task that not block the main agent workflow (#5081)
* feat:为subagent添加后台任务参数

* ruff

* fix: update terminology from 'handoff mission' to 'background task' and refactor related logic

* fix: update terminology from 'background_mission' to 'background_task' in HandoffTool and related logic

* fix(HandoffTool): update background_task description for clarity on usage

---------

Co-authored-by: Soulter <905617992@qq.com>
2026-02-22 13:57:42 +08:00
Soulter
cda49c3a9a fix: remove additionalProperties from tool schema properties (#5253)
fixes: #5217
2026-02-20 16:13:20 +08:00
時壹
3ca8dd204f fix: prevent duplicate error message when all LLM providers fail (#5183) 2026-02-18 16:29:35 +08:00
Soulter
754144ad99 feat: add fallback chat model chain in tool loop runner (#5109)
* feat: implement fallback provider support for chat models and update configuration

* feat: enhance provider selection display with count and chips for selected providers

* feat: update fallback chat providers to use provider settings and add warning for non-list fallback models
2026-02-15 11:51:34 +08:00
Soulter
d2de0ea5ad feat: QQ 官方机器人平台支持主动推送消息、私聊场景下支持接收文件 (#5066)
* feat: QQ 官方机器人平台支持主动推送消息、私聊场景下支持接收文件

* feat: enhance QQOfficialWebhook to remember session scenes for group, channel, and friend messages
2026-02-12 19:07:53 +08:00
Soulter
9d93bda3fe feat: temporary file handling and introduce TempDirCleaner (#5026)
* feat: temporary file handling and introduce TempDirCleaner

- Updated various modules to use `get_astrbot_temp_path()` instead of `get_astrbot_data_path()` for temporary file storage.
- Renamed temporary files for better identification and organization.
- Introduced `TempDirCleaner` to manage the size of the temporary directory, ensuring it does not exceed a specified limit by deleting the oldest files.
- Added configuration option for maximum temporary directory size in the dashboard.
- Implemented tests for `TempDirCleaner` to verify cleanup functionality and size management.

* ruff
2026-02-12 01:04:48 +08:00
Li-shi-ling
aab095347f fix: 'HandoffTool' object has no attribute 'agent' (#5005)
* fix: 移动agent的位置到super().__init__之后

* add: 添加一行注释
2026-02-10 11:01:49 +08:00
Dt8333
7dd95d8a59 chore: auto ann fix by ruff (#4903)
* chore: auto fix by ruff

* refactor: 统一修正返回类型注解为 None/bool 以匹配实现

* refactor: 将 _get_next_page 改为异步并移除多余的请求错误抛出

* refactor: 将 get_client 的返回类型改为 object

* style: 为 LarkMessageEvent 的相关方法添加返回类型注解 None

---------

Co-authored-by: Soulter <37870767+Soulter@users.noreply.github.com>
2026-02-09 00:22:24 +08:00
鸦羽
eca3ede7b0 fix: dedupe preset messages (#4961) 2026-02-08 22:18:13 +08:00
Gao Jinzhe
952023db30 feat: 允许 LLM 预览工具返回的图片并自主决定是否发送 (#4895)
* feat: 允许 LLM 预览工具返回的图片并自主决定是否发送

* 复用 send_message_to_user 替代独立的图片发送工具

* feat: implement _HandleFunctionToolsResult class for improved tool response handling

* docs: add path handling guidelines to AGENTS.md

---------

Co-authored-by: Soulter <905617992@qq.com>
2026-02-08 13:16:16 +08:00
Soulter
4658998b85 fix: messages[x] assistant content must contain at least one part (#4928)
* fix: messages[x] assistant content must contain at least one part

fixes: #4876

* ruff format
2026-02-07 00:33:07 +08:00
boushi1111
d69592aaa8 fix: TypeError when MCP schema type is a list (#4867)
* Fix TypeError when MCP schema type is a list

Fixes crash in Gemini native tools with VRChat MCP.

* Refactor: avoid modifying schema in place per feedback

* Fix formatting and cleanup comments
2026-02-05 22:51:29 +08:00