learn

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Learn Error Pattern Skill

错误模式学习Skill

Operator Context

技能操作上下文

This skill operates as an operator for manual knowledge ingestion, configuring Claude to parse user-provided error-solution pairs and store them in the cross-session learning database. It implements a Parse-Classify-Store-Confirm workflow with deterministic Python scripts performing all database operations.

本技能作为手动知识注入的操作符，配置Claude以解析用户提供的错误-解决方案配对，并将其存储到跨会话学习数据库中。它实现了解析-分类-存储-确认的工作流，所有数据库操作均由确定性Python脚本执行。

Hardcoded Behaviors (Always Apply)

硬编码行为（始终生效）

Input Validation: Never store empty error patterns or empty solutions
Environment Variable Injection Prevention: Always pass values via environment variables to the Python script, never inline user strings into code
High Confidence: Manually taught patterns always receive confidence 0.9
Single Pattern Per Invocation: Store exactly one pattern per
```
/learn
```
call

输入验证：绝不存储空错误模式或空解决方案
防止环境变量注入：始终通过环境变量向Python脚本传递值，绝不将用户字符串直接嵌入代码
高置信度：手动传授的模式始终获得0.9的置信度
每次调用仅存储单个模式：每个
```
/learn
```
调用仅存储一个模式

Default Behaviors (ON unless disabled)

默认行为（启用状态，除非手动禁用）

Auto-Classification: Determine error type and fix type from solution text
Confirmation Output: Display stored pattern details back to user
Duplicate Handling: If pattern already exists, update its confidence to 0.9

自动分类：根据解决方案文本确定错误类型和修复类型
确认输出：向用户显示已存储的模式详情
重复处理：如果模式已存在，将其置信度更新为0.9

Optional Behaviors (OFF unless enabled)

可选行为（禁用状态，除非手动启用）

Batch Mode: Accept multiple patterns from a file (one per line,
```
error -> solution
```
format)
Dry Run: Parse and classify without storing, showing what would be saved

批量模式：从文件中接收多个模式（每行一个，格式为
```
error -> solution
```
）
试运行模式：仅解析和分类，不存储，显示将要保存的内容

What This Skill CAN Do

本技能可实现的功能

Parse "error -> solution" input into structured pattern data
Classify fix types (auto, skill, agent, manual) from solution text
Store patterns in the unified learning database at 0.9 confidence
Update confidence on existing patterns for the same error signature

将「error -> solution」输入解析为结构化模式数据
根据解决方案文本分类修复类型（自动、skill、agent、手动）
以0.9的置信度将模式存储到统一学习数据库中
针对相同错误签名的现有模式更新其置信度

What This Skill CANNOT Do

本技能不可实现的功能

Debug live issues (use systematic-debugging instead)
Automatically learn from tool errors (that is the error-learner hook)

Query or list existing patterns (use

python3 ~/.claude/scripts/learning-db.py query --category error

)

Delete or reset the database (remove
```
~/.claude/learning/learning.db
```
)
Store patterns without user-provided error AND solution text

调试实时问题（请使用systematic-debugging技能）
从工具错误中自动学习（该功能由error-learner钩子负责）

查询或列出现有模式（请使用

python3 ~/.claude/scripts/learning-db.py query --category error

）

删除或重置数据库（请删除
```
~/.claude/learning/learning.db
```
文件）
在用户未同时提供错误和解决方案文本的情况下存储模式

Instructions

操作步骤

Phase 1: PARSE

阶段1：解析

Goal: Extract error pattern and solution from user input.

Step 1: Identify input format

Accepted formats:

```
/learn "error pattern" -> "solution"
```
```
/learn "error pattern" => "solution"
```
Freeform: "teach that X means Y" or "remember: when X, do Y"

Step 2: Extract fields

```
error_pattern
```
: The error message or symptom text
```
solution
```
: The fix or resolution text

Gate: Both error_pattern and solution are non-empty strings. If either is missing, ask the user for the missing part. Do not proceed with empty fields.

目标：从用户输入中提取错误模式和解决方案。

步骤1：识别输入格式

支持的格式：

```
/learn "错误模式" -> "解决方案"
```
```
/learn "错误模式" => "解决方案"
```
自由格式：「告知X对应Y」或「记住：当出现X时执行Y」

步骤2：提取字段

```
error_pattern
```
：错误消息或症状文本
```
solution
```
：修复或解决方法文本

校验门：

error_pattern

和

solution

都必须是非空字符串。如果任一为空，请向用户索要缺失的部分。不得处理空字段。

Phase 2: CLASSIFY

阶段2：分类

Goal: Determine fix type and action from the solution text.

Apply these rules in order:

Solution contains install command (pip install, npm install, apt install) ->
```
fix_type=auto
```
,
```
fix_action=install_dependency
```

Solution contains

replace_all

fix_type=auto

fix_action=use_replace_all

Solution references a skill name ->
```
fix_type=skill
```
,
```
fix_action=<skill-name>
```
Solution references an agent name ->
```
fix_type=agent
```
,
```
fix_action=<agent-name>
```

Otherwise ->

fix_type=manual

fix_action=apply_suggestion

Gate: fix_type and fix_action are determined. Proceed.

目标：根据解决方案文本确定修复类型和操作。

按以下顺序应用规则：

解决方案包含安装命令（pip install、npm install、apt install）→
```
fix_type=auto
```
，
```
fix_action=install_dependency
```

解决方案包含

replace_all

→

fix_type=auto

，

fix_action=use_replace_all

解决方案引用技能名称 →
```
fix_type=skill
```
，
```
fix_action=<skill-name>
```
解决方案引用agent名称 →
```
fix_type=agent
```
，
```
fix_action=<agent-name>
```

其他情况 →

fix_type=manual

，

fix_action=apply_suggestion

校验门：确定fix_type和fix_action后，继续执行。

Phase 3: STORE

阶段3：存储

Goal: Persist the pattern to the learning database.

Execute the storage script using the

learning-db.py

CLI:

bash

python3 ~/.claude/scripts/learning-db.py record \
  "<error_type>" \
  "<error_signature>" \
  "<error_pattern> → <solution>" \
  --category error \
  --confidence 0.9

Where

<error_type>

is the classified type (e.g., "missing_file", "multiple_matches") and

<error_signature>

is a kebab-case key derived from the error pattern.

Example:

bash

python3 ~/.claude/scripts/learning-db.py record \
  "multiple_matches" \
  "edit-tool-multiple-matches" \
  "Edit tool fails with 'found N matches' → Use replace_all=True parameter" \
  --category error \
  --confidence 0.9

Gate: Script exits 0 and prints confirmation. If script fails, see Error Handling.

目标：将模式持久化到学习数据库中。

使用

learning-db.py

CLI执行存储脚本：

bash

python3 ~/.claude/scripts/learning-db.py record \
  "<error_type>" \
  "<error_signature>" \
  "<error_pattern> → <solution>" \
  --category error \
  --confidence 0.9

其中

<error_type>

是分类后的错误类型（例如"missing_file"、"multiple_matches"），

<error_signature>

是从错误模式衍生的短横线命名格式的键。

示例：

bash

python3 ~/.claude/scripts/learning-db.py record \
  "multiple_matches" \
  "edit-tool-multiple-matches" \
  "Edit tool fails with 'found N matches' → Use replace_all=True parameter" \
  --category error \
  --confidence 0.9

校验门：脚本返回0并打印确认信息。如果脚本执行失败，请查看错误处理部分。

Phase 4: CONFIRM

阶段4：确认

Goal: Report stored pattern back to the user.

Display:

Learned pattern:
  Error: "<error_pattern>"
  Solution: "<solution>"
  Type: <fix_type> (<fix_action>)
  Confidence: 0.9

Gate: User sees confirmation. Skill complete.

目标：向用户报告已存储的模式。

显示内容：

已学习模式：
  错误："<error_pattern>"
  解决方案："<solution>"
  类型：<fix_type> (<fix_action>)
  置信度：0.9

校验门：用户看到确认信息后，技能执行完成。

Error Handling

错误处理

Error: "Script fails with ImportError or FileNotFoundError"

错误："Script fails with ImportError or FileNotFoundError"

Cause: scripts/learning-db.py not found or not synced to ~/.claude/scripts/ Solution: Verify working directory is the repo root, or use

~/.claude/scripts/learning-db.py

for cross-repo access.

原因：scripts/learning-db.py未找到或未同步到~/.claude/scripts/目录解决方案：确认工作目录为仓库根目录，或使用

~/.claude/scripts/learning-db.py

以实现跨仓库访问。

Error: "Database locked"

错误："Database locked"

Cause: Another process holds the SQLite lock Solution: Retry after 2 seconds. If persistent, check for hung processes with

lsof ~/.claude/learning/learning.db

原因：另一个进程持有SQLite锁解决方案：2秒后重试。如果问题持续，使用

lsof ~/.claude/learning/learning.db

检查是否有挂起的进程。

Error: "User provides only error, no solution"

错误："User provides only error, no solution"

Cause: Incomplete input Solution: Ask the user explicitly for the solution text. Do not guess or fabricate solutions.

原因：输入不完整解决方案：明确向用户索要解决方案文本。不得猜测或编造解决方案。

Anti-Patterns

反模式

Anti-Pattern 1: Storing Vague Patterns

反模式1：存储模糊模式

What it looks like:

/learn "it broke" -> "fix it"

Why wrong: Pattern too vague to match future errors. Solution provides no actionable guidance. Do instead: Ask user to provide the specific error message and concrete fix steps.

表现：

/learn "it broke" -> "fix it"

问题：模式过于模糊，无法匹配未来的错误。解决方案未提供可操作的指导。 正确做法：请用户提供具体的错误消息和明确的修复步骤。

Anti-Pattern 2: Inlining User Strings Into Python Code

反模式2：将用户字符串嵌入Python代码

What it looks like: Building Python code with f-strings containing user input Why wrong: Injection risk. Quotes or special characters in error text break the script. Do instead: Always pass via environment variables as shown in Phase 3.

表现：使用包含用户输入的f-string构建Python代码问题：存在注入风险。错误文本中的引号或特殊字符会导致脚本崩溃。 正确做法：始终如阶段3所示，通过环境变量传递用户输入。

Anti-Pattern 3: Storing Without Confirming Back

反模式3：存储后不向用户确认

What it looks like: Running the script and saying "done" without showing what was stored Why wrong: User cannot verify correctness. Typos go unnoticed. Do instead: Always complete Phase 4 with the full confirmation output.

表现：运行脚本后仅告知"完成"，不显示存储的内容问题：用户无法验证正确性。输入错误无法被发现。 正确做法：始终完成阶段4，显示完整的确认输出。

References

参考资料

```
hooks/lib/learning_db_v2.py
```
: Unified learning database module
```
scripts/learning-db.py
```
: CLI for recording, querying, and managing learnings
```
hooks/error-learner.py
```
: Automatic error learning hook (complementary system)

```
hooks/lib/learning_db_v2.py
```
：统一学习数据库模块
```
scripts/learning-db.py
```
：用于记录、查询和管理学习内容的CLI工具
```
hooks/error-learner.py
```
：自动错误学习钩子（互补系统）