skills-manager

Skills Manager (Smart Router)

Purpose

Context-aware routing to skills and hooks management. Helps you troubleshoot, fine-tune, and manage the automated documentation system.

When Auto-Activated

• Discussing skills activation or hooks
• Creating new skills
• Keywords: skill activation, hook, troubleshoot, fine-tune, keyword, skill-rules.json, create skill, new skill, agentskills
• Debugging why a skill didn't activate or activated incorrectly

🚨 CRITICAL RULES

1. Logs are your friend - Always check .claude/logs/skill-activations.log first
2. Test after changes - Always verify modifications to skill-rules.json work
3. Validate JSON - Use jq to validate skill-rules.json before committing
4. Keywords should be specific - Too-broad keywords cause false positives

📋 Quick Diagnostic Workflow

Problem: Skill Didn't Activate

1. Check logs:

   tail -20 .claude/logs/skill-activations.log
   

2. Check current keywords:

   cat .claude/hooks/skill-rules.json | jq '.skills."SKILL-NAME".promptTriggers.keywords'
   

3. Add missing keyword:

   • Ask Claude: "Add 'KEYWORD' to SKILL-NAME"
   • Or edit .claude/hooks/skill-rules.json directly

4. Test:

   echo '{"prompt":"test prompt"}' | .claude/hooks/skill-activation-prompt.sh
   

Problem: Skill Activated When Shouldn't

1. Identify the trigger - Check logs to see which keyword matched

2. Remove or make more specific:

   • Remove: "Remove 'text' from localization-developer"
   • Make specific: Replace "text" with "localized text"

3. Verify:

   tail .claude/logs/skill-activations.log
   

🎯 Common Tasks

Add a Keyword

Quick: Ask Claude

"Add 'refactoring' to ios-dev-guidelines keywords"

Manual: Edit .claude/hooks/skill-rules.json

"keywords": [
  "swift",
  "refactoring"  // ← Add here
]

Remove a Keyword

Quick: Ask Claude

"Remove 'text' from localization-developer, it's too broad"

Manual: Edit and remove from keywords array

Check What Activated

tail -20 .claude/logs/skill-activations.log

Look for:

✓ Matched: skill-name     # ← This skill activated
No matches found          # ← Nothing activated

Test Activation Manually

echo '{"prompt":"add feature flag"}' | .claude/hooks/skill-activation-prompt.sh

Should output skill suggestion if match found.

Auto-Learning Feature

What it does: When a substantial prompt (100+ chars OR 3+ lines) doesn't activate any skills, the system prompts you with available skills and auto-updates keywords based on your feedback.

Workflow:

1. You submit a substantial prompt
2. No skills activate
3. System shows: "Should any of these skills be activated?"
4. You respond: "Yes, localization-developer should activate"
5. Claude extracts keywords from your prompt
6. Claude runs: .claude/hooks/utils/add-keywords-to-skill.sh localization-developer <keywords>
7. skill-rules.json updated
8. Future similar prompts auto-activate

Manual keyword extraction:

# Test keyword extraction
echo "Update space settings localization for membership tiers" | .claude/hooks/utils/extract-keywords.sh
# Output: membership, settings, localization, tiers, update

# Add keywords manually
.claude/hooks/utils/add-keywords-to-skill.sh localization-developer "membership" "tiers"

Logs:

• Missed activations: .claude/logs/skill-activations-missed.log
• Learning updates: .claude/logs/skill-learning.log

🔧 The System Components

Hooks (Automation)

Location: .claude/hooks/

What they do:

• skill-activation-prompt.sh - Suggests skills based on your prompt
• post-tool-use-tracker.sh - Tracks file edits
• swiftformat-post-edit.sh - Auto-formats Swift files immediately after edit
• notification-alert.sh - Sends macOS notifications when Claude needs input

Skills (Routers)

Location: .claude/skills/*/SKILL.md

The 7 skills:

1. ios-dev-guidelines - Swift/iOS patterns
2. localization-developer - Localization
3. code-generation-developer - Feature flags, make generate
4. design-system-developer - Icons, typography, colors
5. skills-manager - This skill (meta!)
6. code-review-developer - Code review standards
7. feature-toggle-developer - Feature toggle removal, cleanup detection

Configuration

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

What it contains:

• Keywords for each skill
• Intent patterns (regex)
• File path patterns
• Priority settings

📊 Configuration Structure

{
  "skills": {
    "skill-name": {
      "type": "domain",
      "priority": "high",  // or "medium", "low"
      "description": "Smart router to...",
      "promptTriggers": {
        "keywords": ["keyword1", "keyword2"],
        "intentPatterns": ["(create|add).*?something"]
      },
      "fileTriggers": {
        "pathPatterns": ["**/*.swift"],
        "contentPatterns": ["SomePattern"]
      }
    }
  },
  "config": {
    "maxSkillsPerPrompt": 2,
    "logActivations": true
  }
}

📝 Creating New Skills (agentskills.io Spec)

Skills must follow the agentskills.io specification.

Required SKILL.md Format

Every skill must start with YAML frontmatter:

---
name: skill-name
description: What this skill does and when to use it (max 1024 chars).
---

# Skill Title

## Purpose
...

Name Field Rules

• 1-64 characters
• Lowercase only: a-z, 0-9, -
• No start/end hyphens: skill-name ✅, -skill- ❌
• No consecutive hyphens: my-skill ✅, my--skill ❌
• Must match directory name: skills/my-skill/SKILL.md → name: my-skill

Description Field

• 1-1024 characters
• Must describe: What it does AND when to use it
• Include keywords that help agents identify relevant tasks

Good example:

description: Context-aware routing to the iOS localization system. Use when working with .xcstrings files, Loc constants, or user-facing text.

Optional Frontmatter Fields

---
name: skill-name
description: Required description.
license: MIT
compatibility: Designed for Claude Code
metadata:
  author: anytype
  version: "1.0"
---

Directory Structure

skill-name/
├── SKILL.md          # Required - main skill file
├── scripts/          # Optional - executable scripts
├── references/       # Optional - additional docs
└── assets/           # Optional - templates, images

Quick Checklist for New Skills

• Directory name matches name field
• YAML frontmatter with name and description
• Name is lowercase with hyphens only
• Description explains what AND when
• SKILL.md under 500 lines (use references/ for details)
• Added to skill-rules.json with keywords
• Tested activation with sample prompts

⚠️ Common Issues

Hook Not Executing

Symptom: No activation messages, empty logs

Fix:

# Make hooks executable
chmod +x .claude/hooks/*.sh

# Verify
ls -l .claude/hooks/*.sh  # Should show rwx

Invalid JSON

Symptom: Hook fails silently

Fix:

# Validate
jq . .claude/hooks/skill-rules.json

# If error, ask Claude to fix it

Too Many False Positives

Symptom: Skill activates too often

Fix: Make keywords more specific

• ❌ "text" → activates for everything
• ✅ "localized text" → more specific

📚 Complete Documentation

Full Guide: .claude/SKILLS_MANAGEMENT_GUIDE.md

For comprehensive coverage of:

• Detailed troubleshooting workflows
• Advanced regex patterns for intent matching
• Creating new skills from scratch
• Monitoring and maintenance
• Log rotation and cleanup
• Complete skill-rules.json reference
• Examples for every scenario

✅ Health Check Commands

# Check hook permissions
ls -l .claude/hooks/*.sh

# Validate configuration
jq . .claude/hooks/skill-rules.json

# View recent activations
tail -20 .claude/logs/skill-activations.log

# Count activations by skill
grep "Matched:" .claude/logs/skill-activations.log | sort | uniq -c

# Test specific prompt
echo '{"prompt":"your test"}' | .claude/hooks/skill-activation-prompt.sh

💡 Pro Tips

1. Ask Claude to check logs - "Check the logs for my last prompt"
2. Use logs for tuning - Review regularly to spot patterns
3. Start broad - Add broad keywords, narrow if too many false positives
4. Test everything - Always test after modifying skill-rules.json
5. Document changes - Add comments in skill-rules.json

🔗 Related Docs

• .claude/hooks/README.md - Complete hooks documentation
• .claude/skills/README.md - Skills system overview
• CLAUDE.md - Main documentation

---

Navigation: This is a smart router. For deep troubleshooting and management details, always refer to SKILLS_MANAGEMENT_GUIDE.md.

Quick help: Just ask "Check skills system health" or "Why didn't X skill activate?"