1.3 Progressive Disclosure
Progressive Disclosure
Claude reads only what it needs, when it needs it
Skills use a layered approach so Claude reads only what is relevant. This design keeps Claude fast and efficient — it does not load everything at once, just what matters for the current task.
Three Levels of Disclosure
YAML Frontmatter
Metadata at the top of SKILL.md. Contains name, description, and trigger words. Claude reads this first to decide whether to activate the skill.
SKILL.md Body
The main instructions. Step-by-step workflow, rules, and output format. Claude reads this when the skill is activated.
Linked Files
References, templates, and examples in the skill folder. Claude reads these only when needed during execution.
How It Works in Practice
Level 1: Discovery
Claude scans YAML frontmatter across all skills to find matching ones. Fast, lightweight, no wasted tokens.
Level 2: Activation
Once matched, Claude reads the full SKILL.md body. Gets the complete workflow and rules.
Level 3: Deep Dive
During execution, Claude reads referenced files only when a step requires additional knowledge.
Performance Benefit
This three-tier system means Claude never loads unnecessary content. A skill with 10 reference files only reads the ones actually needed for each specific task.
Design Principle
Progressive disclosure keeps Claude fast and efficient. Each layer adds detail only when needed. This is the same principle used in good UI design — show the right information at the right time.
Tři úrovně informací — Claude čte jen to, co potřebuje.
There are no comments for now.