Skill Capture

Skill 捕获

This skill extracts valuable learnings, patterns, and workflows from conversations and persists them as reusable skill files for future sessions.

此Skill可从对话中提取有价值的经验、模式和工作流，并将其保存为可复用的Skill文件，供后续会话使用。

When This Skill Activates

此Skill的激活场景

Activate when the user mentions ANY of these:

当用户提及以下任意内容时激活：

Explicit Requests

明确请求

"let's save this as a skill"
"this workflow should be reusable"
"capture this as a skill"
"make this a skill"
"turn this into a skill"
"save this pattern"

"我们把这个保存为Skill吧"
"这个工作流应该可以复用"
"把这个捕获为Skill"
"把这个做成Skill"
"把这个转换成Skill"
"保存这个模式"

Workflow Completion Signals

工作流完成信号

Complex problem solved after multiple iterations
Multi-step procedure successfully executed
Valuable domain knowledge discovered through research
Code pattern emerged that could benefit other projects

经过多次迭代后解决了复杂问题
成功执行多步骤流程
通过研究发现有价值的领域知识
出现可用于其他项目的代码模式

Learning Moments

学习时刻

"this was tricky to figure out"
"glad we finally got this working"
"I wish I knew this earlier"
"this should be documented"

"这个真难搞懂"
"很高兴我们终于搞定了"
"我要是早点知道这个就好了"
"这个应该被记录下来"

Capture Process

捕获流程

Follow these 5 phases in order:

按顺序遵循以下5个阶段：

Phase 1: Identification

阶段1：识别

Review the conversation to identify capturable content:

What to Look For:

Category	Example
Workflows	Multi-step procedures that took iterations to perfect
Domain Knowledge	Information requiring research or expertise
Problem Solutions	Approaches that resolved complex issues
Code Patterns	Reusable patterns that could help other projects
Decision Rationale	Architectural choices with clear reasoning

Questions to Ask:

Is this generalizable beyond this specific project?
Would another Claude instance benefit from this knowledge?
Did this require multiple iterations to get right?
Is this non-obvious procedural knowledge?

回顾对话以确定可捕获的内容：

需要寻找的内容：

类别	示例
工作流	经过多次迭代完善的多步骤流程
领域知识	需要研究或专业知识的信息
问题解决方案	解决复杂问题的方法
代码模式	可用于其他项目的复用模式
决策依据	有明确理由的架构选择

需要询问的问题：

这是否能推广到该特定项目之外？
其他Claude实例是否会从这些知识中受益？
这是否需要多次迭代才能正确完成？
这是否是非显而易见的流程知识？

Phase 2: Destination Planning

阶段2：目标规划

Determine where the skill should live:

1. Check if an existing skill should be updated
   - Glob for skills/*/SKILL.md
   - Read related skills to check for overlap

2. If creating new skill:
   - Choose descriptive kebab-case name
   - Create in skills/<skill-name>/ directory

Skill vs. Rule Decision:

Create a Skill	Capture a Rule
Workflow with multiple steps	Single preference statement
Procedural knowledge	"always/never do X"
Domain expertise	Style/format preferences
Code patterns with context	Import/naming conventions

If it's a rule/preference, delegate to

rules-capture

skill instead.

确定Skill的存储位置：

1. 检查是否应更新现有Skill
   - 遍历 skills/*/SKILL.md
   - 阅读相关Skill以检查是否存在重叠

2. 如果创建新Skill：
   - 选择描述性的短横线命名法（kebab-case）名称
   - 在 skills/<skill-name>/ 目录下创建

Skill与规则的决策：

创建Skill	捕获规则
多步骤工作流	单一偏好声明
流程知识	"始终/绝不做X"
领域专业知识	样式/格式偏好
带上下文的代码模式	导入/命名约定

如果是规则/偏好，请转而使用

rules-capture

Skill。

Phase 3: Content Drafting

阶段3：内容起草

Transform conversational insights into SKILL.md format:

markdown

---
name: <skill-name>
description: <One sentence describing what the skill does and when to use it>
version: 1.0.0
tags:
  - <relevant>
  - <tags>
---

将会话见解转换为SKILL.md格式：

markdown

---
name: <skill-name>
description: <一句话描述此Skill的功能及使用场景>
version: 1.0.0
tags:
  - <相关标签>
  - <相关标签>
---

<Skill Title>

<Skill标题>

<Skill用途的简要描述>

When to Use

使用场景

<明确的激活触发条件>

Process

流程

<Step-by-step workflow or procedure>

<分步工作流或流程>

Examples

示例

<展示Skill实际应用的具体示例>

Integration

集成

undefined

<此Skill如何与其他Skill配合使用（如适用）>

undefined

Phase 4: Distillation

阶段4：提炼

Extract and refine the content:

Extract the Final Approach - Capture only what worked, not the failed attempts
Generalize - Remove project-specific details that don't apply broadly
Add Context - Explain why certain approaches work, not just what to do
Include Examples - Add concrete examples that demonstrate usage

Checklist:

Removed project-specific paths and names
Generalized any hardcoded values
Added explanatory context for non-obvious steps
Included both good and bad examples where helpful
Kept language in imperative form

提取并优化内容：

提取最终方法 - 仅捕获有效的内容，而非失败的尝试
通用化 - 删除不适用于广泛场景的项目特定细节
添加上下文 - 解释某些方法有效的原因，而不只是说明要做什么
包含示例 - 添加展示用法的具体示例

检查清单：

移除了项目特定的路径和名称
通用化了所有硬编码值
为非显而易见的步骤添加了解释性上下文
酌情包含了好的和坏的示例
采用命令式语言（“执行X”而非“你应该执行X”）

Phase 5: Verification

阶段5：验证

Ensure the skill is ready for use:

Quality Checks:

Check	Requirement
Length	Under 500 lines (use references/ for large content)
Frontmatter	Valid YAML with name and description
Completeness	All sections have meaningful content
Actionability	Instructions are specific enough to follow
Formatting	Follows library patterns (see rules-capture for reference)

Validation:

bash

undefined

确保Skill已准备好使用：

质量检查：

检查项	要求
长度	少于500行（大内容使用 references/ 目录）
前置元数据	包含名称和描述的有效YAML
完整性	所有部分都有有意义的内容
可操作性	说明足够具体，可遵循
格式	遵循库模式（参考rules-capture）

验证：

bash

undefined

Run the package script to validate

运行包脚本进行验证

scripts/package_skill.py skills/<skill-name>

---

scripts/package_skill.py skills/<skill-name>

---

What to Capture

要捕获的内容

Good Candidates

合适的候选内容

Workflows that took multiple iterations to get right
Domain knowledge that required research
Problem-solving approaches that worked
Code patterns that could be reused across projects
Decision rationale for architectural choices
Integration patterns between tools or services
Debugging approaches for specific technologies

经过多次迭代才完善的工作流
需要研究的领域知识
有效的问题解决方法
可跨项目复用的代码模式
架构选择的决策依据
工具或服务之间的集成模式
特定技术的调试方法

Anti-Patterns (Do NOT Capture)

反模式（请勿捕获）

One-off solutions specific to a single project
Trivial fixes or obvious patterns
Information already documented elsewhere
Sensitive or proprietary information
Incomplete workflows still being refined
Preferences without procedural value (use rules-capture instead)

特定于单个项目的一次性解决方案
微不足道的修复或明显的模式
已在其他地方记录的信息
敏感或专有信息
仍在完善中的不完整工作流
无流程价值的偏好（改用rules-capture）

Output Structure

输出结构

Create two files for each skill:

为每个Skill创建两个文件：

SKILL.md

See Phase 3 template above. Key principles:

Use imperative form ("Do X" not "You should do X")
Be specific about triggers and activation conditions
Include concrete examples
Reference any bundled resources

见阶段3的模板。核心原则：

采用命令式语言（“执行X”而非“你应该执行X”）
明确说明触发条件和激活场景
包含具体示例
引用任何捆绑的资源

plugin.json

json

{
  "name": "<skill-name>",
  "version": "1.0.0",
  "description": "<Same description as SKILL.md frontmatter>",
  "author": {
    "name": "Ship Shit Dev",
    "email": "hello@shipshit.dev",
    "url": "https://github.com/shipshitdev"
  },
  "license": "MIT",
  "skills": "."
}

json

{
  "name": "<skill-name>",
  "version": "1.0.0",
  "description": "<与SKILL.md前置元数据中的描述相同>",
  "author": {
    "name": "Ship Shit Dev",
    "email": "hello@shipshit.dev",
    "url": "https://github.com/shipshitdev"
  },
  "license": "MIT",
  "skills": "."
}

Integration

集成

This skill works with:

Skill	Integration
`rules-capture`	Delegates preference capture; handles skill creation
`session-documenter`	Skills captured during session are noted
`skill-creator`	Uses creation patterns and validation
`docs`	For formatting help with SKILL.md

此Skill可与以下Skill配合使用：

Skill	集成方式
`rules-capture`	委托偏好捕获；处理Skill创建
`session-documenter`	记录会话中捕获的Skill
`skill-creator`	使用创建模式和验证
`docs`	提供SKILL.md的格式帮助

Example Interaction

示例交互

Context: User spent 30 minutes debugging a NestJS queue integration issue. The solution involved specific configuration patterns.

User: "This was a pain to figure out. Let's save this for next time."

Claude:

I'll capture this as a skill. Let me analyze what we learned:

Capturable Knowledge:
- NestJS Bull queue configuration pattern
- Redis connection handling for queues
- Processor registration requirements

This would make a good standalone skill or could extend `nestjs-queue-architect`.

Let me check the existing skill...
[Reads skills/nestjs-queue-architect/SKILL.md]

The existing skill covers queue setup but not this specific debugging pattern.
I'll add a new "Troubleshooting" section.

Created: skills/nestjs-queue-architect/SKILL.md (updated)
- Added troubleshooting section
- Documented Redis connection requirements
- Added common error patterns

Should I also create session documentation for this debugging session?

背景：用户花费30分钟调试NestJS队列集成问题。解决方案涉及特定的配置模式。

用户："这个真难搞，我们把这个保存下来下次用吧。"

Claude：

我会把这个捕获为Skill。让我分析一下我们学到的内容：

可捕获的知识：
- NestJS Bull队列配置模式
- 队列的Redis连接处理
- 处理器注册要求

这可以做成一个独立的Skill，或者扩展`nestjs-queue-architect`。

让我检查现有的Skill...
[读取 skills/nestjs-queue-architect/SKILL.md]

现有Skill涵盖了队列设置，但未包含这个特定的调试模式。
我会添加一个新的“故障排除”部分。

已创建：skills/nestjs-queue-architect/SKILL.md（已更新）
- 添加了故障排除部分
- 记录了Redis连接要求
- 添加了常见错误模式

我是否也需要为这次调试会话创建会话文档？

Activation Triggers

激活触发器

The skill activates on these patterns:

regex

undefined

此Skill在以下模式下激活：

regex

undefined

Explicit capture requests

明确的捕获请求

Workflow completion signals

工作流完成信号

Learning signals

学习信号

(wish I knew|should have known|next time|for future)


---

**When this skill is active**, automatically identify valuable knowledge from the conversation and offer to persist it as reusable skills for future sessions.

(wish I knew|should have known|next time|for future)


---

**当此Skill处于激活状态时**，会自动从对话中识别有价值的知识，并主动提出将其保存为可复用的Skill，供后续会话使用。

skill-capture

Original

Translation

Skill Capture

Skill 捕获

When This Skill Activates

此Skill的激活场景

Explicit Requests

明确请求

Workflow Completion Signals

工作流完成信号

Learning Moments

学习时刻

Capture Process

捕获流程

Phase 1: Identification

阶段1：识别

Phase 2: Destination Planning

阶段2：目标规划

Phase 3: Content Drafting

阶段3：内容起草

<Skill Title>

<Skill标题>

When to Use

使用场景

Process

流程

Examples

示例

Integration

集成

Phase 4: Distillation

阶段4：提炼

Phase 5: Verification

阶段5：验证

Run the package script to validate

运行包脚本进行验证

What to Capture

要捕获的内容

Good Candidates

合适的候选内容

Anti-Patterns (Do NOT Capture)

反模式（请勿捕获）

Output Structure

输出结构

SKILL.md

SKILL.md

plugin.json

plugin.json

Integration

集成

Example Interaction

示例交互

Activation Triggers

激活触发器

Explicit capture requests

明确的捕获请求

Workflow completion signals

工作流完成信号

Learning signals

学习信号