articleJanuary 2, 2026

Claude Code Skills Don't Auto-Activate

Claude Code skills fail to auto-activate despite documentation claims; hook-based workarounds achieve only 40-50% reliability, so manual invocation remains the practical choice.

The Problem

Claude Code's documentation claims skills are "model-invoked" and autonomously activated. In practice, they consistently fail to engage without explicit intervention. Even when user requests match skill descriptions exactly, Claude ignores them.

Attempted Workaround

Hook scripts can inject explicit instructions into prompts via ~/.claude/settings.json. The key distinction:

  • Ineffective: Gentle reminders like "Check skills"
  • Effective: Direct commands like "Use Skill(research) to handle this request"

Reality Check

Testing revealed only 40-50% reliability even with explicit hooks. The workaround is essentially unreliable. Manual skill invocation remains the practical choice when accuracy matters.

Connections (22)