skill-factory

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Skill Factory

Autonomous skill creation - just tell me what you need, I'll handle everything.

自主技能创建——只需告知你的需求，其余交由我处理。

What This Does

功能介绍

You request a skill, I deliver a production-ready skill with guaranteed quality (score >= 8.0/10).

No decision-making required. No tool selection. No quality checking. Just results.

你提出技能需求，我将交付质量有保障（评分≥8.0/10）的可生产级技能。

无需用户决策，无需选择工具，无需手动质检，直接获取结果。

Anthropic's Official Best Practices

Anthropic官方最佳实践

For comprehensive guidance on creating effective skills, see:

references/overview.md - Complete overview of Agent Skills architecture, progressive disclosure, and how Skills work across different platforms (API, Claude Code, Agent SDK, claude.ai)
references/quickstart.md - Quick tutorial on using pre-built Agent Skills in the Claude API with practical code examples
references/best-practices.md - Detailed authoring best practices including core principles, skill structure, progressive disclosure patterns, workflows, evaluation strategies, and common patterns
references/anthropic-best-practices.md - Quality scoring system (10/10 criteria) used by skill-factory

These references provide Anthropic's official guidance and are consulted during the quality assurance phase.

如需创建有效技能的全面指导，请查看：

references/overview.md - Agent Skills架构、渐进式披露的完整概述，以及技能在不同平台（API、Claude Code、Agent SDK、claude.ai）的工作方式
references/quickstart.md - 在Claude API中使用预构建Agent Skills的快速教程，包含实用代码示例
references/best-practices.md - 详细的创作最佳实践，包括核心原则、技能结构、渐进式披露模式、工作流、评估策略和常见模式
references/anthropic-best-practices.md - skill-factory使用的质量评分系统（10分制标准）

这些参考资料提供了Anthropic的官方指导，会在质量保证阶段被查阅。

Usage

使用方法

Simply describe the skill you need:

"Create a skill for Anchor development with latest docs and best practices"
"Create a React skill from react.dev with comprehensive examples"
"Create a skill for Solana transaction debugging workflows"
"Create a skill for writing technical documentation following company standards"

I will automatically:

✅ Analyze your request
✅ Select the optimal creation method
✅ Create the skill
✅ Run quality assurance loops (until score >= 8.0)
✅ Test with automated scenarios
✅ Deliver ready-to-use skill with stats

只需描述你需要的技能：

"Create a skill for Anchor development with latest docs and best practices"
"Create a React skill from react.dev with comprehensive examples"
"Create a skill for Solana transaction debugging workflows"
"Create a skill for writing technical documentation following company standards"

我会自动完成以下步骤：

✅ 分析你的需求
✅ 选择最优创建方式
✅ 创建技能
✅ 运行质量保证循环（直到评分≥8.0）
✅ 通过自动化场景测试
✅ 交付可直接使用的技能及统计数据

What You Get

交付成果示例

✅ anchor-development skill ready!

📊 Quality Score: 8.9/10 (Excellent)
📝 Lines: 412 (using progressive disclosure)
📚 Coverage: 247 documentation pages
💡 Examples: 68 code samples
🧪 Test Pass Rate: 100% (15/15 scenarios)

📁 Location: ~/.claude/skills/anchor-development/
📦 Zip: ~/Downloads/anchor-development.zip

Try it: "How do I create an Anchor program?"

✅ anchor-development skill ready!

📊 Quality Score: 8.9/10 (Excellent)
📝 Lines: 412 (using progressive disclosure)
📚 Coverage: 247 documentation pages
💡 Examples: 68 code samples
🧪 Test Pass Rate: 100% (15/15 scenarios)

📁 Location: ~/.claude/skills/anchor-development/
📦 Zip: ~/Downloads/anchor-development.zip

Try it: "How do I create an Anchor program?"

How It Works (Behind the Scenes)

工作原理（幕后流程）

Phase 1: Request Analysis (Automatic)

第一阶段：需求分析（自动完成）

I analyze your request to determine:

Source Detection:

Documentation URL/mention? → Automated scraping path
"Latest docs", "current version"? → Automated path
GitHub repository mention? → Automated path
PDF/manual path? → Automated path
Custom workflow/process description? → Manual TDD path
Both documentation AND custom needs? → Hybrid path

Quality Requirements Extraction:

"Best practices" → Enforce quality gates
"Latest version" → Scrape current docs
"Examples" → Ensure code samples included
"Comprehensive" → Verify coverage completeness

我会分析你的需求以确定：

来源检测：

是否提及文档URL？→ 自动爬取路径
是否提到“最新文档”“当前版本”？→ 自动路径
是否提及GitHub仓库？→ 自动路径
是否为PDF/手动路径？→ 自动路径
是否为自定义工作流/流程描述？→ 手动TDD路径
是否同时包含文档和自定义需求？→ 混合路径

质量要求提取：

“最佳实践”→ 启用质量关卡
“最新版本”→ 爬取当前文档
“示例”→ 确保包含代码示例
“全面覆盖”→ 验证内容覆盖完整性

Phase 2: Execution (Automatic)

第二阶段：执行（自动完成）

Path A: Documentation-Based (Skill_Seekers)

Detected: Documentation source available
Method: Automated scraping with quality enhancement

Steps I take:
1. Check Skill_Seekers installation (install if needed)
2. Configure scraping parameters automatically
3. Run scraping with optimal settings
4. Monitor progress
5. Initial quality check
6. If score < 8.0: Run enhancement loop
7. Re-score until >= 8.0
8. Test with auto-generated scenarios
9. Package and deliver

Path B: Custom Workflows (Manual TDD)

Detected: Custom workflow/process
Method: Test-Driven Documentation (obra methodology)

Steps I take:
1. Create pressure test scenarios
2. Run baseline (without skill)
3. Document agent behavior
4. Write minimal skill addressing baseline
5. Test with skill present
6. Identify rationalizations/gaps
7. Close loopholes
8. Iterate until bulletproof
9. Package and deliver

Path C: Hybrid

Detected: Documentation + custom requirements
Method: Scrape then enhance

Steps I take:
1. Scrape documentation (Path A)
2. Identify gaps vs requirements
3. Fill gaps with TDD approach (Path B)
4. Unify and test as whole
5. Quality loop until >= 8.0
6. Package and deliver

路径A：基于文档（Skill_Seekers）

Detected: Documentation source available
Method: Automated scraping with quality enhancement

Steps I take:
1. Check Skill_Seekers installation (install if needed)
2. Configure scraping parameters automatically
3. Run scraping with optimal settings
4. Monitor progress
5. Initial quality check
6. If score < 8.0: Run enhancement loop
7. Re-score until >= 8.0
8. Test with auto-generated scenarios
9. Package and deliver

路径B：自定义工作流（手动TDD）

Detected: Custom workflow/process
Method: Test-Driven Documentation (obra methodology)

Steps I take:
1. Create pressure test scenarios
2. Run baseline (without skill)
3. Document agent behavior
4. Write minimal skill addressing baseline
5. Test with skill present
6. Identify rationalizations/gaps
7. Close loopholes
8. Iterate until bulletproof
9. Package and deliver

路径C：混合方式

Detected: Documentation + custom requirements
Method: Scrape then enhance

Steps I take:
1. Scrape documentation (Path A)
2. Identify gaps vs requirements
3. Fill gaps with TDD approach (Path B)
4. Unify and test as whole
5. Quality loop until >= 8.0
6. Package and deliver

Phase 3: Quality Assurance Loop (Automatic)

第三阶段：质量保证循环（自动完成）

I enforce Anthropic best practices:

python

while quality_score < 8.0:
    issues = analyze_against_anthropic_guidelines(skill)

    if "vague_description" in issues:
        improve_description_specificity()

    if "missing_examples" in issues:
        extract_or_generate_examples()

    if "too_long" in issues:
        apply_progressive_disclosure()

    if "poor_structure" in issues:
        reorganize_content()

    quality_score = rescore()

Quality Criteria (Anthropic Best Practices):

✅ Description: Specific, clear, includes when_to_use
✅ Conciseness: <500 lines OR progressive disclosure
✅ Examples: Concrete code samples, not abstract
✅ Structure: Well-organized, clear sections
✅ Name: Follows conventions (lowercase, hyphens, descriptive)

Important: The quality assurance process consults references/best-practices.md for Anthropic's complete authoring guidelines and references/anthropic-best-practices.md for the 10-point scoring criteria.

我会严格执行Anthropic最佳实践：

python

while quality_score < 8.0:
    issues = analyze_against_anthropic_guidelines(skill)

    if "vague_description" in issues:
        improve_description_specificity()

    if "missing_examples" in issues:
        extract_or_generate_examples()

    if "too_long" in issues:
        apply_progressive_disclosure()

    if "poor_structure" in issues:
        reorganize_content()

    quality_score = rescore()

质量标准（Anthropic最佳实践）：

✅ 描述：具体清晰，包含when_to_use字段
✅ 简洁性：少于500行或采用渐进式披露
✅ 示例：具体代码示例，而非抽象描述
✅ 结构：组织有序，章节清晰
✅ 命名：遵循规范（小写、连字符、描述性命名）

重要提示：质量保证过程会参考references/best-practices.md获取Anthropic完整的创作指南，并参考references/anthropic-best-practices.md的10分制评分标准。

Phase 4: Testing (Automatic)

第四阶段：测试（自动完成）

I generate and run test scenarios:

python

undefined

我会生成并运行测试场景：

python

undefined

Auto-generate test cases from skill content

test_cases = extract_key_topics(skill)

for topic in test_cases: query = f"How do I {topic}?"

# Test WITHOUT skill (baseline)
baseline = run_query_without_skill(query)

# Test WITH skill
with_skill = run_query_with_skill(query)

# Verify improvement
if not is_better(with_skill, baseline):
    identify_gap()
    enhance_skill()
    retest()

undefined

test_cases = extract_key_topics(skill)

for topic in test_cases: query = f"How do I {topic}?"

# Test WITHOUT skill (baseline)
baseline = run_query_without_skill(query)

# Test WITH skill
with_skill = run_query_with_skill(query)

# Verify improvement
if not is_better(with_skill, baseline):
    identify_gap()
    enhance_skill()
    retest()

undefined

Phase 5: Delivery (Automatic)

第五阶段：交付（自动完成）

Package skill:
- Create skill directory structure
- Generate SKILL.md with frontmatter
- Create reference files (if using progressive disclosure)
- Add examples directory
- Create .zip for easy upload
- Install to ~/.claude/skills/ (if desired)
- Generate summary statistics

Package skill:
- Create skill directory structure
- Generate SKILL.md with frontmatter
- Create reference files (if using progressive disclosure)
- Add examples directory
- Create .zip for easy upload
- Install to ~/.claude/skills/ (if desired)
- Generate summary statistics

Progress Reporting

进度报告

You'll see real-time progress:

🔍 Analyzing request...
   ✅ Detected: Documentation-based (docs.rs/anchor-lang)
   ✅ Requirements: Latest version, best practices, examples

🔄 Creating skill...
   📥 Scraping docs.rs/anchor-lang... (2 min)
   📚 Extracting 247 pages...
   💾 Organizing content...

📊 Quality check: 7.4/10
   ⚠️  Issues found:
       - Description too generic (fixing...)
       - Missing examples in 4 sections (adding...)
       - Some outdated patterns (updating...)

🔧 Enhancing skill...
   ✏️  Description improved
   📝 Examples added
   🔄 Patterns updated

📊 Quality check: 8.9/10 ✅

🧪 Testing...
   ✅ 15/15 scenarios passing

✅ anchor-development skill ready!

你将看到实时进度：

🔍 Analyzing request...
   ✅ Detected: Documentation-based (docs.rs/anchor-lang)
   ✅ Requirements: Latest version, best practices, examples

🔄 Creating skill...
   📥 Scraping docs.rs/anchor-lang... (2 min)
   📚 Extracting 247 pages...
   💾 Organizing content...

📊 Quality check: 7.4/10
   ⚠️  Issues found:
       - Description too generic (fixing...)
       - Missing examples in 4 sections (adding...)
       - Some outdated patterns (updating...)

🔧 Enhancing skill...
   ✏️  Description improved
   📝 Examples added
   🔄 Patterns updated

📊 Quality check: 8.9/10 ✅

🧪 Testing...
   ✅ 15/15 scenarios passing

✅ anchor-development skill ready!

Dependencies

依赖项

Required:

Python 3.10+ (for quality scripts)
bash (for automation scripts)

Optional (auto-installed when needed):

Skill_Seekers (for documentation scraping)
- Will prompt for installation on first documentation-based request
- One-command setup:
```
scripts/install-skill-seekers.sh
```

必需：

Python 3.10+（用于质量脚本）
bash（用于自动化脚本）

可选（需要时自动安装）：

Skill_Seekers（用于文档爬取）
- 首次处理基于文档的需求时会提示安装
- 一键安装：
```
scripts/install-skill-seekers.sh
```

Configuration

配置

Environment variables (optional):

bash

undefined

环境变量（可选）：

bash

undefined

Skill_Seekers installation path

export SKILL_SEEKERS_PATH="$HOME/Skill_Seekers"

Output directory for created skills

export SKILL_OUTPUT_DIR="$HOME/.claude/skills"

Minimum quality score (default: 8.0)

export MIN_QUALITY_SCORE="8.0"

Auto-install dependencies (default: prompt)

export AUTO_INSTALL_DEPS="true"

undefined

export AUTO_INSTALL_DEPS="true"

undefined

Examples

示例

Example 1: Documentation Skill

User: "Create a React skill from react.dev"

Agent:
🔍 Analyzing... → Documentation-based
🔄 Scraping react.dev... → 3 min
📊 Quality: 7.8 → 8.6 ✅
🧪 Testing: 12/12 passing ✅
✅ react-development skill ready (8.6/10)

Example 2: Custom Workflow Skill

User: "Create a skill for debugging Solana transaction failures"

Agent:
🔍 Analyzing... → Custom workflow (no docs to scrape)
📝 Using TDD methodology...
🧪 RED: Testing baseline...
✏️  GREEN: Writing skill...
🔄 REFACTOR: Closing loopholes...
📊 Quality: 8.3 ✅
✅ solana-transaction-debugging skill ready (8.3/10)

Example 3: Hybrid Skill

User: "Create an Anchor skill with docs plus custom debugging workflows"

Agent:
🔍 Analyzing... → Hybrid (docs + custom)
📥 Scraping docs.rs/anchor-lang... → 2 min
📝 Adding custom debugging workflows...
🔄 Integrating and testing...
📊 Quality: 8.9 ✅
✅ anchor-development skill ready (8.9/10)

示例1：基于文档的技能

User: "Create a React skill from react.dev"

Agent:
🔍 Analyzing... → Documentation-based
🔄 Scraping react.dev... → 3 min
📊 Quality: 7.8 → 8.6 ✅
🧪 Testing: 12/12 passing ✅
✅ react-development skill ready (8.6/10)

示例2：自定义工作流技能

User: "Create a skill for debugging Solana transaction failures"

Agent:
🔍 Analyzing... → Custom workflow (no docs to scrape)
📝 Using TDD methodology...
🧪 RED: Testing baseline...
✏️  GREEN: Writing skill...
🔄 REFACTOR: Closing loopholes...
📊 Quality: 8.3 ✅
✅ solana-transaction-debugging skill ready (8.3/10)

示例3：混合方式技能

User: "Create an Anchor skill with docs plus custom debugging workflows"

Agent:
🔍 Analyzing... → Hybrid (docs + custom)
📥 Scraping docs.rs/anchor-lang... → 2 min
📝 Adding custom debugging workflows...
🔄 Integrating and testing...
📊 Quality: 8.9 ✅
✅ anchor-development skill ready (8.9/10)

Quality Guarantee

质量保证

Every skill delivered by skill-factory:

✅ Scores >= 8.0/10 on Anthropic best practices
✅ Has concrete examples (not abstract)
✅ Follows structure conventions
✅ Tested with auto-generated scenarios
✅ Ready to use immediately

If quality < 8.0, I keep working until it reaches 8.0+

Skill Factory交付的每一项技能：

✅ 符合Anthropic最佳实践，评分≥8.0/10
✅ 包含具体示例，而非抽象描述
✅ 遵循结构规范
✅ 经过自动生成场景的测试
✅ 可直接投入使用

如果评分低于8.0，我会持续优化直至达到8.0以上

Troubleshooting

故障排除

Skill_Seekers installation fails:

bash

undefined

Skill_Seekers安装失败：

bash

undefined

Manual installation

git clone https://github.com/yusufkaraaslan/Skill_Seekers ~/Skill_Seekers cd ~/Skill_Seekers pip install -r requirements.txt

Or use installation script

~/Projects/claude-skills/skill-factory/skill/scripts/install-skill-seekers.sh


**Quality score stuck below 8.0:**
- I'll report what's blocking and suggest manual review
- Check references/anthropic-best-practices.md for criteria
- Run manual enhancement if needed

**Want to understand methodology:**
- See references/obra-tdd-methodology.md (testing approach)
- See references/anthropic-best-practices.md (quality criteria)
- See references/skill-seekers-integration.md (automation details)

~/Projects/claude-skills/skill-factory/skill/scripts/install-skill-seekers.sh


**质量评分卡在8.0以下：**
- 我会报告阻碍因素并建议手动检查
- 查看references/anthropic-best-practices.md了解评分标准
- 必要时可进行手动优化

**想了解方法论细节：**
- 查看references/obra-tdd-methodology.md（测试方法）
- 查看references/anthropic-best-practices.md（质量标准）
- 查看references/skill-seekers-integration.md（自动化细节）

Reference Files

参考文件

Anthropic Official Documentation:

references/overview.md - Agent Skills architecture, progressive disclosure, and platform details
references/quickstart.md - Quick tutorial on using pre-built Agent Skills in the Claude API
references/best-practices.md - Comprehensive authoring guidelines from Anthropic
references/anthropic-best-practices.md - Quality scoring system (10/10 criteria)

Skill Factory Implementation Details:

references/obra-tdd-methodology.md - Full TDD testing approach
references/skill-seekers-integration.md - Automation documentation
references/request-analysis.md - How requests are parsed
references/quality-loops.md - Enhancement algorithms

Anthropic官方文档：

references/overview.md - Agent Skills架构、渐进式披露及平台细节
references/quickstart.md - 在Claude API中使用预构建Agent Skills的快速教程
references/best-practices.md - Anthropic提供的全面创作指南
references/anthropic-best-practices.md - 10分制质量评分系统

Skill Factory实现细节：

references/obra-tdd-methodology.md - 完整TDD测试方法
references/skill-seekers-integration.md - 自动化文档
references/request-analysis.md - 需求解析方式
references/quality-loops.md - 优化算法

Scripts Reference

脚本参考

Available helper scripts in

scripts/

directory:

check-skill-seekers.sh - Check if Skill_Seekers is installed
install-skill-seekers.sh - One-command Skill_Seekers setup
quality-check.py - Score any skill against Anthropic best practices

Usage examples:

bash

undefined

scripts/

目录下的可用辅助脚本：

check-skill-seekers.sh - 检查Skill_Seekers是否已安装
install-skill-seekers.sh - 一键安装Skill_Seekers
quality-check.py - 根据Anthropic最佳实践为任意技能评分

使用示例：

bash

undefined

Check Skill_Seekers installation

./scripts/check-skill-seekers.sh

Install Skill_Seekers

./scripts/install-skill-seekers.sh

Quality check a skill

python3 ./scripts/quality-check.py /path/to/skill/SKILL.md

undefined

python3 ./scripts/quality-check.py /path/to/skill/SKILL.md

undefined

Philosophy

设计理念

You don't want to:

Navigate decision trees
Choose between tools
Check quality manually
Test with subagents yourself
Wonder if output is good

You want to:

Describe what you need
Get high-quality result
Start using immediately

That's what skill-factory delivers.

你不想做的事：

浏览决策树
在工具间做选择
手动检查质量
自行使用子Agent测试
不确定输出质量是否达标

你想做的事：

描述你的需求
获取高质量结果
立即投入使用

这正是Skill Factory所能提供的。

Credits

致谢

Built on top of excellent tools:

Skill_Seekers - Documentation scraping
obra/superpowers-skills - TDD methodology
Anthropic skill-creator - Best practices

Skill-factory orchestrates these tools with automatic quality assurance and testing.

Just tell me what skill you need. I'll handle the rest.

基于以下优秀工具构建：

Skill_Seekers - 文档爬取
obra/superpowers-skills - TDD方法论
Anthropic skill-creator - 最佳实践

Skill Factory通过自动质量保证和测试，将这些工具进行编排整合。

只需告诉我你需要什么技能，其余交给我处理。