smart-skill-loading

What problem does it solve?

This Skill dramatically reduces AI token usage and context bloat by intelligently loading only the most relevant skill information for a given task. This leads to significant cost savings and a 10% improvement in AI response accuracy by providing targeted context.

Core Features & Use Cases

• Metadata-First Loading: Loads only lightweight YAML frontmatter from all skills initially, saving up to 1400 tokens from the system prompt.
• Conditional Full Loading: Only loads the full SKILL.md content for skills that are highly relevant to the current task, based on keyword matching.
• Progressive Disclosure: Supporting files (examples, templates) are loaded on-demand, further optimizing token usage.
• Use Case: When managing a large set of AI skills, this pattern ensures that only the "PDF extraction" skill is fully loaded for a task like "Extract fields from invoice.pdf," avoiding unnecessary context from other skills and reducing processing costs.

Quick Start

Example: Smartly load skills based on a task description

This function would be part of your agent's core logic.

def load_skills_smart(task_description, skills_dir): # ... (implementation details for metadata scan, matching, conditional loading) ... return loaded_skills

User asks for a PDF task, only relevant skills are loaded

loaded_skills = load_skills_smart("Extract fields from invoice.pdf", ".claude/skills/")

Result: Only the 'progressive-metadata' skill (if relevant) would be fully loaded.