claude skill activation hook

check_skills.sh

#!/bin/bash

# Hook: UserPromptSubmit
# Purpose: ENFORCE skill checking before starting work

# Return JSON with proper hookSpecificOutput structure per Claude Code docs
cat <<'EOF'
{
  "decision": undefined ,
  "reason": "n/a",
  "hookSpecificOutput": {
    "hookEventName": "UserPromptSubmit",
    "additionalContext": "🚨 MANDATORY: Before doing ANY work, you MUST create a todo list with TodoWrite containing:\n\nTodo 1 (status: pending): \"Check available skills for matches\"\nTodo 2 (status: pending): \"[User's actual request]\"\n\nYou MUST complete Todo 1 FIRST by:\n1. Check if ANY of your available skills (listed in your system prompt under Available Skills) match the user's request by comparing their descriptions to the request\n2. If one or more skills match → Mark Todo 1 complete, then INVOKE ALL matching skills using the Skill tool (invoke multiple if needed). Do NOT proceed to Todo 2 when skills match.\n3. If NO skills match → Mark Todo 1 complete, then proceed to Todo 2\n\nIMPORTANT: Use the skill descriptions already in your system prompt. Do NOT read filesystem or glob for skills. You already know what skills are available.\n\nYou are REQUIRED to use TodoWrite before starting. You are FORBIDDEN from skipping Todo 1."
  }
}
EOF

@michabbb lol what do you mean 50%?

Oh I see what you mean lol