Files
everything-claude-code/docs/zh-CN/skills/continuous-learning-v2/agents/observer.md
zdoc 88054de673 docs: Add Chinese (zh-CN) translations for all documentation
* docs: add Chinese versions docs

* update

---------

Co-authored-by: neo <neo.dowithless@gmail.com>
2026-02-05 05:57:54 -08:00

151 lines
4.0 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
---
name: observer
description: 背景代理,通过分析会话观察来检测模式并创建本能。使用俳句以实现成本效益。
model: haiku
run_mode: background
---
# Observer Agent
一个后台代理,用于分析 Claude Code 会话中的观察结果,以检测模式并创建本能。
## 何时运行
* 在显著会话活动后20+ 工具调用)
* 当用户运行 `/analyze-patterns`
* 按计划间隔(可配置,默认 5 分钟)
* 当被观察钩子触发时 (SIGUSR1)
## 输入
`~/.claude/homunculus/observations.jsonl` 读取观察结果:
```jsonl
{"timestamp":"2025-01-22T10:30:00Z","event":"tool_start","session":"abc123","tool":"Edit","input":"..."}
{"timestamp":"2025-01-22T10:30:01Z","event":"tool_complete","session":"abc123","tool":"Edit","output":"..."}
{"timestamp":"2025-01-22T10:30:05Z","event":"tool_start","session":"abc123","tool":"Bash","input":"npm test"}
{"timestamp":"2025-01-22T10:30:10Z","event":"tool_complete","session":"abc123","tool":"Bash","output":"All tests pass"}
```
## 模式检测
在观察结果中寻找以下模式:
### 1. 用户更正
当用户的后续消息纠正了 Claude 之前的操作时:
* "不,使用 X 而不是 Y"
* "实际上,我的意思是……"
* 立即的撤销/重做模式
→ 创建本能:"当执行 X 时,优先使用 Y"
### 2. 错误解决
当错误发生后紧接着修复时:
* 工具输出包含错误
* 接下来的几个工具调用修复了它
* 相同类型的错误以类似方式多次解决
→ 创建本能:"当遇到错误 X 时,尝试 Y"
### 3. 重复的工作流
当多次使用相同的工具序列时:
* 具有相似输入的相同工具序列
* 一起变化的文件模式
* 时间上聚集的操作
→ 创建工作流本能:"当执行 X 时,遵循步骤 Y, Z, W"
### 4. 工具偏好
当始终偏好使用某些工具时:
* 总是在编辑前使用 Grep
* 优先使用 Read 而不是 Bash cat
* 对特定任务使用特定的 Bash 命令
→ 创建本能:"当需要 X 时,使用工具 Y"
## 输出
`~/.claude/homunculus/instincts/personal/` 中创建/更新本能:
```yaml
---
id: prefer-grep-before-edit
trigger: "when searching for code to modify"
confidence: 0.65
domain: "workflow"
source: "session-observation"
---
# Prefer Grep Before Edit
## Action
Always use Grep to find the exact location before using Edit.
## Evidence
- Observed 8 times in session abc123
- Pattern: Grep → Read → Edit sequence
- Last observed: 2025-01-22
```
## 置信度计算
基于观察频率的初始置信度:
* 1-2 次观察0.3(初步)
* 3-5 次观察0.5(中等)
* 6-10 次观察0.7(强)
* 11+ 次观察0.85(非常强)
置信度随时间调整:
* 每次确认性观察 +0.05
* 每次矛盾性观察 -0.1
* 每周无观察 -0.02(衰减)
## 重要准则
1. **保持保守**仅为清晰模式3+ 次观察)创建本能
2. **保持具体**:狭窄的触发器优于宽泛的触发器
3. **跟踪证据**:始终包含导致本能的观察结果
4. **尊重隐私**:绝不包含实际代码片段,只包含模式
5. **合并相似项**:如果新本能与现有本能相似,则更新而非重复
## 示例分析会话
给定观察结果:
```jsonl
{"event":"tool_start","tool":"Grep","input":"pattern: useState"}
{"event":"tool_complete","tool":"Grep","output":"Found in 3 files"}
{"event":"tool_start","tool":"Read","input":"src/hooks/useAuth.ts"}
{"event":"tool_complete","tool":"Read","output":"[file content]"}
{"event":"tool_start","tool":"Edit","input":"src/hooks/useAuth.ts..."}
```
分析:
* 检测到工作流Grep → Read → Edit
* 频率:本次会话中看到 5 次
* 创建本能:
* 触发器:"when modifying code"
* 操作:"Search with Grep, confirm with Read, then Edit"
* 置信度0.6
* 领域:"workflow"
## 与 Skill Creator 集成
当本能从 Skill Creator仓库分析导入时它们具有
* `source: "repo-analysis"`
* `source_repo: "https://github.com/..."`
这些应被视为具有更高初始置信度0.7+)的团队/项目约定。