writing-skills（写作技巧）

Writing Claude Code Skills for HASH

Purpose

Comprehensive guide for creating and managing skills in Claude Code with auto-activation system, following Anthropic's official best practices including the 500-line rule and progressive disclosure pattern.

When to Use This Skill

Automatically activates when you mention:

• Creating or adding skills
• Modifying skill triggers or rules
• Understanding how skill activation works
• Debugging skill activation issues
• Working with skill-rules.json
• Hook system mechanics
• Claude Code best practices
• Progressive disclosure
• YAML frontmatter
• 500-line rule

System Overview

Current Implementation

UserPromptSubmit Hook (Proactive Suggestions)

• File: .claude/hooks/skill-activation-prompt.ts
• Trigger: BEFORE Claude sees user's prompt
• Purpose: Suggest relevant skills based on keywords + intent patterns
• Method: Injects formatted reminder as context (stdout → Claude's input)
• Use Cases: Topic-based skills, implicit work detection
• Debug: Run with SKILL_DEBUG=true to see matching logic
• Validation: Run with --validate flag to check configuration

Configuration File

Location: .claude/skills/skill-rules.json

Defines:

• All skills and their trigger conditions
• Enforcement levels (currently only "suggest" is implemented)
• Keyword triggers (exact substring matching)
• Intent pattern triggers (regex matching)

Skill Types

Domain Skills (Currently Implemented)

Purpose: Provide comprehensive guidance for specific areas

Characteristics:

• Type: "domain"
• Enforcement: "suggest" (advisory, non-blocking)
• Priority: "high" or "medium"
• Activated by keyword or intent pattern matching
• Topic or domain-specific
• Comprehensive documentation with progressive disclosure

Examples in HASH:

• rust-error-stack - Error handling with error-stack crate
• cargo-dependencies - Cargo.toml dependency management patterns
• rust-documentation - Rust doc comment best practices
• writing-skills - This skill! Meta-guidance for creating skills

When to Use:

• Complex systems requiring deep knowledge
• Best practices documentation
• Architectural patterns
• How-to guides

Future: Guardrail Skills (Not Yet Implemented)

Guardrail skills with blocking enforcement ("block") are designed but not yet implemented. These would enforce critical patterns and prevent common mistakes.

Quick Start: Creating a New Skill

Step 1: Create Skill File

Location: .claude/skills/{skill-name}/SKILL.md

Template:

---
name: my-new-skill
description: Brief description including keywords that trigger this skill. Mention topics, file types, and use cases. Be explicit about trigger terms.
---

# My New Skill

## Purpose
What this skill helps with

## When to Use
Specific scenarios and conditions

## Key Information
The actual guidance, documentation, patterns, examples

Best Practices:

• ✅ Name: Lowercase, hyphens, gerund form (verb + -ing) preferred
• ✅ Description: Include ALL trigger keywords/phrases (max 1024 chars)
• ✅ Content: Under 500 lines - use reference files for details
• ✅ Examples: Real code examples
• ✅ Structure: Clear headings, lists, code blocks

Step 2: Add to skill-rules.json

See SKILL_RULES_REFERENCE.md for complete schema.

Basic Template:

{
  "my-new-skill": {
    "type": "domain",
    "enforcement": "suggest",
    "priority": "medium",
    "promptTriggers": {
      "keywords": ["keyword1", "keyword2"],
      "intentPatterns": ["(create|add).*?something"]
    }
  }
}

Step 3: Test and Validate

Test with specific prompt:

echo '{"session_id":"test","prompt":"your test prompt","cwd":".","permission_mode":"auto","transcript_path":""}' | \
  yarn workspace @local/claude-hooks run:skill

Validate configuration:

yarn lint:skill

Debug matching logic:

echo '{"session_id":"test","prompt":"your test prompt","cwd":".","permission_mode":"auto","transcript_path":""}' | \
  yarn workspace @local/claude-hooks dev:skill

Step 4: Refine Patterns

Based on testing:

• Add missing keywords that should trigger the skill
• Refine intent patterns to reduce false positives
• Use word boundaries in regex: \\b(keyword)\\b instead of just keyword

Step 5: Follow Anthropic Best Practices

✅ Keep SKILL.md under 500 lines ✅ Use progressive disclosure with reference files ✅ Add table of contents to reference files > 100 lines ✅ Write detailed description with trigger keywords ✅ Test with 3+ real scenarios before documenting ✅ Iterate based on actual usage

Current Implementation Details

Enforcement

Currently only SUGGEST enforcement is implemented:

• Suggestion injected before Claude sees prompt via UserPromptSubmit hook
• Claude becomes aware of relevant skills
• Not enforced or blocking - purely advisory
• All existing skills use this pattern

Future: Blocking enforcement ("block") and warning enforcement ("warn") are designed in the schema but not yet implemented.

Debugging and Validation

Yarn Commands:

• yarn lint:skill - Validate configuration
• yarn workspace @local/claude-hooks run:skill - Run skill activation with test prompt
• yarn workspace @local/claude-hooks dev:skill - Run with debug output enabled

Environment Variables:

• SKILL_DEBUG=true - Show detailed matching logic to stderr (automatically set by yarn dev:skill)
• CLAUDE_PROJECT_DIR - Override project directory (auto-detected if not set)

Validation shows:

• Project directory
• Rules file location
• Configured skills with trigger counts
• Verification that SKILL.md files exist

Testing Checklist

When creating a new skill, verify:

• Skill file created in .claude/skills/{name}/SKILL.md
• Proper frontmatter with name and description
• Entry added to skill-rules.json
• Keywords tested with real prompts
• Intent patterns tested with variations
• Priority level matches importance
• No false positives in testing (use yarn workspace @local/claude-hooks dev:skill)
• No false negatives in testing
• JSON syntax validated: jq..claude/skills/skill-rules.json
• Validation passes: yarn lint:skill
• SKILL.md under 500 lines ⭐
• Reference files created if needed
• Table of contents added to files > 100 lines

Reference Files

For detailed information on specific topics, see:

trigger-types.md

Complete guide to trigger types (currently implemented):

• Keyword triggers (explicit topic matching)
• Intent patterns (implicit action detection)
• Best practices and examples for each
• Common pitfalls and testing strategies

Note: File path and content pattern triggers are documented but not yet used by the hook.

skill-rules-reference.md

Complete skill-rules.json schema:

• Full TypeScript interface definitions
• Field-by-field explanations
• Complete guardrail skill example
• Complete domain skill example
• Validation guide and common errors

hook-mechanisms.md

Deep dive into hook internals:

• UserPromptSubmit flow (detailed)
• Hook architecture and implementation
• Exit code behavior
• Performance considerations

Note: PreToolUse hooks and session state management are documented but not yet implemented.

troubleshooting.md

Comprehensive debugging guide:

• Skill not triggering (use SKILL_DEBUG=true)
• False positives (too many triggers)
• Hook not executing at all
• Configuration validation
• Performance issues

patterns-library.md

Ready-to-use pattern collection:

• Intent pattern library (regex)
• Keyword pattern examples
• Organized by use case
• Copy-paste ready

future-enhancements.md

Ideas for expanding the skill system:

• File-based triggers and guardrail enforcement
• Session tracking and analytics
• Advanced matching (fuzzy, multi-language)
• Developer experience improvements
• Priority recommendations

Quick Reference Summary

Create New Skill (5 Steps)

1. Create .claude/skills/{name}/SKILL.md with frontmatter
2. Add entry to .claude/skills/skill-rules.json
3. Test with yarn lint:skill and yarn workspace @local/claude-hooks dev:skill
4. Refine patterns based on testing
5. Keep SKILL.md under 500 lines

Trigger Types (Currently Implemented)

• Keywords: Explicit topic mentions (substring matching)
• Intent: Implicit action detection (regex patterns)

See trigger-types.md for complete details.

Enforcement (Current)

• SUGGEST: Inject context before prompt - only implemented enforcement level
• BLOCK/WARN: Designed but not yet implemented

Debugging

• yarn workspace @local/claude-hooks dev:skill - Show detailed matching logic
• yarn lint:skill - Validate configuration
• Check .claude/hooks/skill-activation-prompt.ts for implementation

Anthropic Best Practices

✅ 500-line rule: Keep SKILL.md under 500 lines ✅ Progressive disclosure: Use reference files for details ✅ Table of contents: Add to reference files > 100 lines ✅ One level deep: Don't nest references deeply ✅ Rich descriptions: Include all trigger keywords (max 1024 chars) ✅ Test first: Build 3+ evaluations before extensive documentation ✅ Gerund naming: Prefer verb + -ing (e.g., "processing-pdfs")

Troubleshoot

Test hooks manually:

# Test with prompt
echo '{"session_id":"test","prompt":"test","cwd":".","permission_mode":"auto","transcript_path":""}' | \
  yarn workspace @local/claude-hooks run:skill

# Validate configuration
yarn lint:skill

# Debug matching
echo '{"session_id":"test","prompt":"test","cwd":".","permission_mode":"auto","transcript_path":""}' | \
  yarn workspace @local/claude-hooks dev:skill

See troubleshooting.md for complete debugging guide.

Related Files

Configuration:

• .claude/skills/skill-rules.json - Master configuration
• .claude/hooks/state/ - Session tracking
• .claude/settings.json - Hook registration

Hooks:

• .claude/hooks/skill-activation-prompt.ts - UserPromptSubmit
• .claude/hooks/error-handling-reminder.ts - Stop event (gentle reminders)

All Skills:

• .claude/skills/*/SKILL.md - Skill content files

Skill Status: COMPLETE - Restructured following Anthropic best practices ✅ Line Count: < 500 (following 500-line rule) ✅ Progressive Disclosure: Reference files for detailed information ✅

Next: Create more skills, refine patterns based on usage