Skill Activation Hook: Guarantee 100% Skill Loading in Claude Code
Stop Claude from ignoring your skills. The Skill Activation Hook appends recommendations to your prompts before Claude sees them. Zero memory reliance.
Problem: You tell Claude Code to load a skill. It forgets. You add instructions to CLAUDE.md. It ignores them. You end up manually reminding Claude about skills that should be automatic.
Quick Win: The Skill Activation Hook intercepts your prompts and appends skill recommendations before Claude sees the message. Claude can't forget because it never had to remember.
When you type "help me implement a feature", Claude actually sees:
Claude now knows exactly which skills to load. No guessing. No forgetting.
How It Works
The hook uses Claude Code's UserPromptSubmit event. Every prompt you send triggers this flow:
- You type a message - Your natural language request
- Hook intercepts - Before Claude sees anything
- Pattern matching - Hook checks
skill-rules.jsonfor keyword and intent matches - Append recommendations - Matching skills get added to your message
- Claude receives both - Your prompt plus skill guidance
The hook runs in milliseconds. You won't notice any delay.
The Matching System
Two strategies work together:
Keyword Matching - Simple string matching. If your prompt contains "commit" or "git push", the git-commits skill triggers.
Intent Patterns - Regex for natural language variation. A pattern like (implement|build).*?feature catches "let's implement this feature" and "build a new feature for me".
Configuration: skill-rules.json
Every skill has triggers defined in .claude/skills/skill-rules.json:
Priority levels control how the hook groups suggestions:
- Critical - Must load before any work
- High - Strongly recommended
- Medium - Helpful context
- Low - Optional enhancement
Customization for Your Speech Patterns
The hook adapts to how you talk. If you always say "push my code" instead of "git push", add it:
After creating any new skill, update its triggers in skill-rules.json. Then read those triggers and keep those speech patterns in mind when prompting.
Session Intelligence
The hook tracks what it already recommended. If it suggested session-management earlier in your conversation, it won't repeat the suggestion. Less noise, same coverage.
Session state lives in recommendation-log.json and auto-cleans after 7 days.
Setup in ClaudeFast
The hook is pre-configured. Verify your .claude/settings.local.json includes:
Windows:
Linux/Mac:
Common Issues
No suggestions appearing Check that your keywords match your actual speech patterns. Run the hook manually to test:
Suggestions appearing when not needed Your keywords may be too broad. Use more specific terms or intent patterns.
Duplicate suggestions The hook might be configured in both global and project settings. Keep it in one location only.
Next Actions
- Check your
skill-rules.jsonmatches your vocabulary - Add keywords for new skills you create
- Learn more about CLAUDE.md configuration to complement the hook
- Explore configuration basics for complete setup
- Review the skills guide if you need to create new skills
The Skill Activation Hook removes human memory from the equation. You focus on describing what you need. The framework handles which skills Claude should use. That's the point of a framework.
Last updated on