Load skills through three progressive disclosure layers to reduce token usage.
npx skills add https://github.com/tody-agent/codymaster --skill cm-skill-index
Please help me install this Agent Skill. Skill: cm-skill-index Source: https://github.com/tody-agent/codymaster/tree/main/skills/cm-skill-index Command: npx skills add https://github.com/tody-agent/codymaster --skill cm-skill-index
Progressive Disclosure accelerates discovery by avoiding full-skill loads, dramatically reducing token usage during skill selection.
Start by enabling Layer 1 discovery and progressively enable Layer 2 and Layer 3 as you decide to execute a skill.
High-intent search queries and answers about installing and using this skill.
Progressive disclosure reduces token waste by indexing skills in a three-layer system, loading only minimal 100-token summaries for initial discovery instead of full skill files. It scopes discovery to quick matching and defers full loads until execution.
Progressive disclosure is a layered skill loading mechanism that works by maintaining a Layer 1 index of names and triggers, a Layer 2 summary for context validation, and a Layer 3 full SKILL.md file loaded only at execution time to minimize upfront token cost.
Implement layered loading by enabling Layer 1 discovery first, then progressively loading Layer 2 context summaries and Layer 3 full instructions only after choosing a candidate skill. This prevents loading unnecessary execution details during scanning.
Yes, you can scan a large catalog using Layer 1 data, which requires only about 100 tokens per skill for fast matching. Layer 2 data uses roughly 300 tokens for closer validation, ensuring full SKILL.md files are never loaded until execution.
Progressive skill indexing validates context using Layer 2 data, which includes descriptions, use cases, and integration notes. This ensures the chosen candidate matches requirements before loading the Layer 3 full SKILL.md at execution time.
The best way to structure skill metadata for token-efficient discovery is a three-layer format: Layer 1 holds 100-token summaries for indexing, Layer 2 contains 300-token context details, and Layer 3 stores the full SKILL.md for execution.
AppsVortexEstablish layered visual testing strategies comparing live UI captures to prototype baselines.
kettleofketchupConfigures shadcn/ui and Tailwind CSS styling layers for React projects.
haidonglethqbOptimize LLM context windows with deterministic token budgeting and probe-based evaluation.
tlancas25Index full skill outputs with FTS5 while returning compact summaries.
