fix: talk_value=0 时真正跳过回复判断，仅保留表达学习

[hsd221]

问题

WebUI 中将群聊 talk_value 配置为 0 后，消息仍会频繁进入 Timing Gate 产生 LLM 调用。根因有三个：

1. _get_effective_reply_frequency() 将频率钳位在 max(0.01, ...)，导致 talk_value=0 实际生效为 0.01
2. 空闲补偿机制 _should_trigger_message_turn_by_idle_compensation() 可将空窗时间折算为等效消息数，绕过阈值
3. get_talk_value() 中 time=""（WebUI 显示「全天」）被 parse_range() 丢弃，导致会话专属规则静默失效。用户将群聊频率设为 0，但因规则被跳过而回退到基础 talk_value=0.8，频率检查 <=0 从未触发

修复

• 移除 _get_effective_reply_frequency() 和 adjust_talk_frequency() 中的 0.01 钳位
• _get_message_trigger_threshold() 频率为 0 时返回 _EFFECTIVELY_INFINITE_THRESHOLD 避免除零
• _schedule_message_turn() 频率为 0 时仅在学习条件满足时调度 turn
• run_loop() 频率为 0 且非强制触发时，学习完成后跳过 Timing Gate 和 Planner
• _is_time_in_range(): time 为空视为全时段生效，修复 WebUI「全天」规则被跳过的问题
• 提取 _is_expression_learning_due() 消除调度与执行阶段的守卫重复
• 提取 _transition_to_idle() 消除 run_loop 两处相同的状态重置代码
• @强制回复不受 talk_value=0 影响

测试

• 现有 12 个 timing gate 测试全部通过
• ruff 检查通过

🤖 Generated with Claude Code

Summary by CodeRabbit

发布说明

• 新功能

  • 新增完全静默模式，当对话频率设置为零时，Agent 可正确进入静默状态。

• 改进

  • 优化消息处理流程，缓存消息检测时跳过不必要的处理步骤，提升效率。
  • 重构对话时间触发逻辑，统一处理时间范围判断。

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 43989c6c-8871-4477-b115-b2d5879ff045

📥 Commits

Reviewing files that changed from the base of the PR and between a0a5b77 and 31da969.

📒 Files selected for processing (3)

• src/common/utils/utils_config.py
• src/maisaka/reasoning_engine.py
• src/maisaka/runtime.py

🚧 Files skipped from review as they are similar to previous changes (2)

• src/maisaka/reasoning_engine.py
• src/maisaka/runtime.py

---

Walkthrough

移除回复频率的非零下限以支持完全静默（有效频率 ≤ 0）；在调度中短路常规回复但保留表达式学习触发；提取聊天规则的时间窗判断为辅助方法；推理引擎在静默且有缓存消息时跳过内部循环并切换空闲。

变更内容

完全静默模式支持与时间窗抽取重构

Layer / File(s)	Summary
常量与频率调整 src/maisaka/runtime.py	添加 _EFFECTIVELY_INFINITE_THRESHOLD 常量；adjust_talk_frequency() 直接保存传入倍数，移除最小非零限制。
频率计算与触发阈值 src/maisaka/runtime.py	_get_effective_reply_frequency() 不再强制非零下限，允许返回 0；_get_message_trigger_threshold() 将有效频率 ≤ 0 映射为 _EFFECTIVELY_INFINITE_THRESHOLD，否则计算 ceil(1.0 / frequency)。
消息调度逻辑 src/maisaka/runtime.py	_schedule_message_turn() 在有效频率 ≤ 0 时短路：仅在表达式学习到期时调度，否则返回跳过正常内部处理。
学习触发逻辑重构 src/maisaka/runtime.py	新增 _is_expression_learning_due() 封装学习触发判断；_trigger_expression_learning() 使用该方法进行早期返回决策。
聊天配置时间窗判定 src/common/utils/utils_config.py	新增 ChatConfigUtils._is_time_in_range(time_str, now_min)，在 get_talk_value 的会话规则与全局回退规则中复用该方法以判断规则时间窗（支持空串与跨午夜）。
推理引擎集成 src/maisaka/reasoning_engine.py	run_loop 添加早期继续守护：当有缓存消息且有效频率 ≤ 0 且无强制计时触发时，记录调试信息并切换到空闲后继续外层循环；finally 路径调用 _transition_to_idle()。

🎯 3 (中等复杂度) | ⏱️ ~20 分钟

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	PR标题准确概括了主要变更：当talk_value=0时跳过回复判断，仅保留表达学习。标题清晰具体，直接反映了此次修复的核心内容。
Description check	✅ Passed	PR描述结构完整，详细说明了问题根因、修复方案和测试情况。相比模板虽未完全填写全部复选框，但核心内容（问题、修复、测试）已充分阐述。
Docstring Coverage	✅ Passed	Docstring coverage is 94.74% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/maisaka/reasoning_engine.py`:
- Around line 456-464: The fast-path early `continue` when cached_messages and
runtime timing conditions are met skips the later cleanup that sets
`_agent_state` back from "running" (which causes `register_message()` to treat
subsequent messages as still-in-progress and incorrectly trigger debounce/quiet
windows); fix by performing the same state reset/phase update before returning
from this branch (or move this timing-gate branch into the shared try/finally
cleanup so the post-run cleanup always runs). Update the branch that checks
`cached_messages` / `_runtime._get_effective_reply_frequency()` /
`_runtime._has_forced_timing_trigger()` to either call the same reset routine
used after the run or ensure `_agent_state` is cleared and any debounce/phase
flags are updated before `continue`, and keep `register_message`/debounce
behaviour consistent.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: db85c620-c075-4648-a18b-b684e310dfd4

📥 Commits

Reviewing files that changed from the base of the PR and between 09eeefc and a0a5b77.

📒 Files selected for processing (2)

• src/maisaka/reasoning_engine.py
• src/maisaka/runtime.py