Skill Search
Enhance Claude Code's skill discovery by replacing static, token-heavy listings with dynamic, semantic retrieval over full skill descriptions.
Enhance Claude Code's skill discovery by replacing static, token-heavy listings with dynamic, semantic retrieval over full skill descriptions.
Skill Search addresses the growing 'token tax' in Claude Code incurred by large skill sets. It replaces the native per-turn injection of all skill descriptions into context with a semantic vector retriever. By setting most skills to 'name-only', the tool frees up significant context window space, providing a measurable reduction in token usage per turn. It also resolves the 'name-bias' issue, allowing Claude to semantically match and invoke relevant skills even when their names don't directly keyword-match the user's query, ensuring more accurate and efficient skill utilization.