feat: enhance persona tool management and update UI localization for subagent orchestration#4990
Conversation
…subagent orchestration
dosubot
bot
added
the
size:M
|
Documentation Updates 1 document(s) were updated by changes in this PR: pr4697的改动View Changes@@ -8,7 +8,7 @@
#### 架构说明
主动代理系统引入了 SubAgent（子代理）架构。主代理（MainAgent）不仅专注于对话和任务委派，还可以直接使用自身工具集。具体工具和任务也可以由子代理（SubAgent）处理。这样可以有效避免 prompt 膨胀和调用失败，同时提升主代理的灵活性。
-- 工具分配：每个 SubAgent 继承自 Persona，拥有独立的工具集、技能、名称和描述。主代理可以直接调用自身工具，也可以通过 `transfer_to_<subagent>` 工具将任务转交给指定子代理。
+- 工具分配：每个 SubAgent 继承自 Persona，拥有独立的工具集、技能、名称和描述。主代理可以直接调用自身工具，也可以通过 `transfer_to_<subagent>` 工具将任务转交给指定子代理。启用 SubAgent 编排后，主代理会保留自身工具，并自动挂载 handoff 工具（transfer_to_*），可根据配置决定是否移除与子代理重复的工具。
- 任务流转：主代理既可以直接处理任务，也可以分发任务给子代理。子代理之间可通过 transfer 工具实现任务转移。主代理和子代理的工具集可以根据配置灵活分配，支持去重和 handoff 工具自动挂载。
#### 使用示例
@@ -24,7 +24,15 @@
这样，主代理既可以自主处理，也可以灵活委派任务。
#### 配置说明
-SubAgent 的定义与 Persona 配置一致，需在配置文件中指定 tools、skills、name、description 等。主代理在未启用 SubAgent 编排时，可直接挂载自身工具集（按 persona 规则，默认全部），并直接调用工具。启用 SubAgent 编排后，主代理仅挂载 transfer_to_* 委派工具，不再直接挂载 persona/全局工具；由 SubAgent 负责工具调用与结果整理。工具分配可通过 UI 或配置文件完成。
+SubAgent 的定义与 Persona 配置一致，需在配置文件中指定 tools、skills、name、description 等。
+
+主代理在未启用 SubAgent 编排时，可直接挂载自身工具集（按 persona 规则，默认全部），并直接调用工具。
+
+启用 SubAgent 编排后，主代理会保留自身工具，并自动挂载 transfer_to_* 委派工具。是否移除与子代理重复的工具由 `remove_main_duplicate_tools` 配置项控制：
+- `remove_main_duplicate_tools: false`（默认）：主代理保留自身工具，即使与子代理重复。
+- `remove_main_duplicate_tools: true`：主代理会移除与子代理重复的工具，仅保留独有工具和 handoff 工具。
+
+工具分配可通过 UI 或配置文件完成。相关配置项说明已在配置文件注释中明确。
### 2. FutureTask（定时任务）机制
@@ -96,7 +104,7 @@
- Persona 选择与工具分配
- 配置持久化
- 供应商类型选择：现在支持在“选择供应商”下拉菜单中选择 `chat_completion` 和 `agent_runner` 类型，便于根据不同需求配置子代理的执行方式。
-- 页面副标题已更新为：“主 LLM 负责聊天与委派（transfer_to_*），工具挂载在各个 SubAgent 上。”
+- 页面副标题已更新为：“主 LLM 可直接使用自身工具，也可通过 handoff 分派给各个 SubAgent。”
主代理和子代理的工具分配逻辑在 UI 中有清晰展示，支持灵活配置。
How did I do? Any feedback? |
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Hey - 我发现了 1 个问题，并留下了一些总体反馈：
- 子代理 orchestrator 代码块里的去重逻辑看起来被反转了：现在在
remove_dup为 true 时，不再是从主工具集里移除交接工具（handoff tools），而是从req.func_tool中移除任何不在handoff_names中的assigned_tools，这很可能会把非交接工具也无意中删掉；建议恢复为基于主工具集名称来检查/移除。 build_main_agent中新增的调试日志会打印完整的prompt、system_prompt、contexts以及工具名称，这可能过于冗长且潜在包含敏感信息；建议对这些字段做截断，或者把这条日志放到更严格的配置/详细等级标志后面。
Prompt for AI Agents
Please address the comments from this code review:
## Overall Comments
- The duplicate-removal logic in the subagent orchestrator block looks inverted: instead of removing handoff tools from the main toolset when `remove_dup` is true, it now removes any `assigned_tools` not in `handoff_names` from `req.func_tool`, which is likely to strip non-handoff tools unintentionally; consider reverting to checking/removing against the main toolset’s names.
- The new debug log in `build_main_agent` prints full `prompt`, `system_prompt`, `contexts`, and tool names, which may be verbose and potentially sensitive; consider truncating these fields or gating this log behind a more restrictive config/verbosity flag.
## Individual Comments
### Comment 1
<location> `astrbot/core/astr_main_agent.py:961-966` </location>
<code_context>
if action_type == "live":
req.system_prompt += f"\n{LIVE_MODE_SYSTEM_PROMPT}\n"
+ logger.debug("Final ProviderRequest: prompt=%s, system_prompt=%s, image_urls=%s, contexts=%s, func_tool=%s",
+ req.prompt,
+ req.system_prompt,
+ req.image_urls,
+ req.contexts,
+ req.func_tool.names() if req.func_tool else None,
+ )
+
</code_context>
<issue_to_address>
**🚨 suggestion (security):** 最终的调试日志可能过于冗长，并可能暴露大量或敏感内容。
在一条调试日志中包含完整的 `prompt`、`system_prompt` 和 `contexts`，可能会泄露用户数据，并生成非常长的日志行，尤其是在生产或高并发环境中。请考虑对这些字段进行截断、根据配置做脱敏处理，或者把这条日志隐藏在更细粒度的调试标志之后，以便在需要时可以关闭完整内容的日志。
建议实现方式：
```python
if action_type == "live":
req.system_prompt += f"\n{LIVE_MODE_SYSTEM_PROMPT}\n"
def _preview_for_logging(value, max_len: int = 512):
"""Return a truncated string representation for safe logging."""
if value is None:
return None
text = str(value)
if len(text) <= max_len:
return text
return text[:max_len] + f"... [truncated {len(text) - max_len} chars]"
logger.debug(
"Final ProviderRequest: prompt_preview=%s, system_prompt_preview=%s, image_urls=%s, contexts_preview=%s, func_tool=%s",
_preview_for_logging(req.prompt),
_preview_for_logging(req.system_prompt),
req.image_urls,
_preview_for_logging(req.contexts),
req.func_tool.names() if req.func_tool else None,
)
# inject toolset in the persona
if (persona and persona.get("tools") is None) or not persona:
```
如果你希望对敏感内容的日志有更严格的控制，可以考虑：
1. 引入一个配置开关（例如 `settings.LOG_FULL_LLM_CONTENT`），并：
- 当该开关为 `False` 时仅使用 `_preview_for_logging`（如上）。
- 只在非生产环境中为调试目的显式开启时，才记录完整内容。
2. 如果其他地方也需要类似的截断逻辑，可以将 `_preview_for_logging` 移动到一个共享的 logging/util 模块中。
</issue_to_address>帮我变得更有用！请在每条评论上点击 👍 或 👎，我会根据你的反馈改进后续评审。
Original comment in English
Hey - I've found 1 issue, and left some high level feedback:
- The duplicate-removal logic in the subagent orchestrator block looks inverted: instead of removing handoff tools from the main toolset when
remove_dupis true, it now removes anyassigned_toolsnot inhandoff_namesfromreq.func_tool, which is likely to strip non-handoff tools unintentionally; consider reverting to checking/removing against the main toolset’s names. - The new debug log in
build_main_agentprints fullprompt,system_prompt,contexts, and tool names, which may be verbose and potentially sensitive; consider truncating these fields or gating this log behind a more restrictive config/verbosity flag.
Prompt for AI Agents
Please address the comments from this code review:
## Overall Comments
- The duplicate-removal logic in the subagent orchestrator block looks inverted: instead of removing handoff tools from the main toolset when `remove_dup` is true, it now removes any `assigned_tools` not in `handoff_names` from `req.func_tool`, which is likely to strip non-handoff tools unintentionally; consider reverting to checking/removing against the main toolset’s names.
- The new debug log in `build_main_agent` prints full `prompt`, `system_prompt`, `contexts`, and tool names, which may be verbose and potentially sensitive; consider truncating these fields or gating this log behind a more restrictive config/verbosity flag.
## Individual Comments
### Comment 1
<location> `astrbot/core/astr_main_agent.py:961-966` </location>
<code_context>
if action_type == "live":
req.system_prompt += f"\n{LIVE_MODE_SYSTEM_PROMPT}\n"
+ logger.debug("Final ProviderRequest: prompt=%s, system_prompt=%s, image_urls=%s, contexts=%s, func_tool=%s",
+ req.prompt,
+ req.system_prompt,
+ req.image_urls,
+ req.contexts,
+ req.func_tool.names() if req.func_tool else None,
+ )
+
</code_context>
<issue_to_address>
**🚨 suggestion (security):** Final debug log may be overly verbose and could expose large or sensitive content.
Including full `prompt`, `system_prompt`, and `contexts` in a single debug log can leak user data and produce very large log lines, especially in production or high-traffic environments. Please consider truncating these fields, redacting them based on configuration, or guarding this behind a more targeted debug flag so full content logging can be disabled where needed.
Suggested implementation:
```python
if action_type == "live":
req.system_prompt += f"\n{LIVE_MODE_SYSTEM_PROMPT}\n"
def _preview_for_logging(value, max_len: int = 512):
"""Return a truncated string representation for safe logging."""
if value is None:
return None
text = str(value)
if len(text) <= max_len:
return text
return text[:max_len] + f"... [truncated {len(text) - max_len} chars]"
logger.debug(
"Final ProviderRequest: prompt_preview=%s, system_prompt_preview=%s, image_urls=%s, contexts_preview=%s, func_tool=%s",
_preview_for_logging(req.prompt),
_preview_for_logging(req.system_prompt),
req.image_urls,
_preview_for_logging(req.contexts),
req.func_tool.names() if req.func_tool else None,
)
# inject toolset in the persona
if (persona and persona.get("tools") is None) or not persona:
```
If you want even stricter control over logging sensitive content, consider:
1. Introducing a configuration flag (e.g. `settings.LOG_FULL_LLM_CONTENT`) and:
- Only use `_preview_for_logging` (as above) when the flag is `False`.
- Log full content only when explicitly enabled for debugging in non-production environments.
2. Moving `_preview_for_logging` to a shared logging/util module if similar truncation is needed elsewhere.
</issue_to_address>Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.
astrbot/core/astr_main_agent.py
Outdated
Comment on lines
961
to
966
| logger.debug("Final ProviderRequest: prompt=%s, system_prompt=%s, image_urls=%s, contexts=%s, func_tool=%s", | ||
| req.prompt, | ||
| req.system_prompt, | ||
| req.image_urls, | ||
| req.contexts, | ||
| req.func_tool.names() if req.func_tool else None, |
Contributor
There was a problem hiding this comment.
🚨 suggestion (security): 最终的调试日志可能过于冗长，并可能暴露大量或敏感内容。
在一条调试日志中包含完整的 prompt、system_prompt 和 contexts，可能会泄露用户数据，并生成非常长的日志行，尤其是在生产或高并发环境中。请考虑对这些字段进行截断、根据配置做脱敏处理，或者把这条日志隐藏在更细粒度的调试标志之后，以便在需要时可以关闭完整内容的日志。
建议实现方式：
if action_type == "live":
req.system_prompt += f"\n{LIVE_MODE_SYSTEM_PROMPT}\n"
def _preview_for_logging(value, max_len: int = 512):
"""Return a truncated string representation for safe logging."""
if value is None:
return None
text = str(value)
if len(text) <= max_len:
return text
return text[:max_len] + f"... [truncated {len(text) - max_len} chars]"
logger.debug(
"Final ProviderRequest: prompt_preview=%s, system_prompt_preview=%s, image_urls=%s, contexts_preview=%s, func_tool=%s",
_preview_for_logging(req.prompt),
_preview_for_logging(req.system_prompt),
req.image_urls,
_preview_for_logging(req.contexts),
req.func_tool.names() if req.func_tool else None,
)
# inject toolset in the persona
if (persona and persona.get("tools") is None) or not persona:如果你希望对敏感内容的日志有更严格的控制，可以考虑：
- 引入一个配置开关（例如
settings.LOG_FULL_LLM_CONTENT），并：- 当该开关为
False时仅使用_preview_for_logging（如上）。 - 只在非生产环境中为调试目的显式开启时，才记录完整内容。
- 当该开关为
- 如果其他地方也需要类似的截断逻辑，可以将
_preview_for_logging移动到一个共享的 logging/util 模块中。
Original comment in English
🚨 suggestion (security): Final debug log may be overly verbose and could expose large or sensitive content.
Including full prompt, system_prompt, and contexts in a single debug log can leak user data and produce very large log lines, especially in production or high-traffic environments. Please consider truncating these fields, redacting them based on configuration, or guarding this behind a more targeted debug flag so full content logging can be disabled where needed.
Suggested implementation:
if action_type == "live":
req.system_prompt += f"\n{LIVE_MODE_SYSTEM_PROMPT}\n"
def _preview_for_logging(value, max_len: int = 512):
"""Return a truncated string representation for safe logging."""
if value is None:
return None
text = str(value)
if len(text) <= max_len:
return text
return text[:max_len] + f"... [truncated {len(text) - max_len} chars]"
logger.debug(
"Final ProviderRequest: prompt_preview=%s, system_prompt_preview=%s, image_urls=%s, contexts_preview=%s, func_tool=%s",
_preview_for_logging(req.prompt),
_preview_for_logging(req.system_prompt),
req.image_urls,
_preview_for_logging(req.contexts),
req.func_tool.names() if req.func_tool else None,
)
# inject toolset in the persona
if (persona and persona.get("tools") is None) or not persona:If you want even stricter control over logging sensitive content, consider:
- Introducing a configuration flag (e.g.
settings.LOG_FULL_LLM_CONTENT) and:- Only use
_preview_for_logging(as above) when the flag isFalse. - Log full content only when explicitly enabled for debugging in non-production environments.
- Only use
- Moving
_preview_for_loggingto a shared logging/util module if similar truncation is needed elsewhere.
dosubot
bot
added
area:webui
5 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
fix: #4971
Modifications / 改动点
Screenshots or Test Results / 运行截图或测试结果
Checklist / 检查清单
requirements.txt和pyproject.toml文件相应位置。/ I have ensured that no new dependencies are introduced, OR if new dependencies are introduced, they have been added to the appropriate locations inrequirements.txtandpyproject.toml.Summary by Sourcery
优化人物（persona）工具分配与子代理（subagent）编排行为，并让仪表盘 UI 和本地化与更新后的子代理流程保持一致。
改进点：
ProviderRequest的详细调试日志记录，以便排查提示词、上下文及工具选择行为相关问题。Original summary in English
Summary by Sourcery
Refine persona tool assignment and subagent orchestration behavior while aligning dashboard UI and localization with the updated subagent flow.
Enhancements: