Claude Code glossary
Progressive disclosure
Progressive disclosure loads skill content on demand, reducing initial context cost.
Also called: progressive disclosure skills, lazy loading skills
Progressive disclosure is the loading model that makes skills cheap. In a regular session, only skill SKILL.md's description and when_to_use text are loaded into context so Claude knows what is available; the full skill content loads only when the skill is invoked.
A third tier exists: skills can include multiple files in their directory, and those reference files load only when Claude actually reads them. So there are three tiers: description always in context, SKILL.md body on invocation, and bundled files on demand.
The combined description and when_to_use text is truncated at 1,536 characters in the skill listing to reduce context usage (configurable via skillListingMaxDescChars). Once a skill loads, its content stays in context across turns, so every line of the body is a recurring token cost. Keep the SKILL.md body concise and push long reference material into separate files.
Subagents with preloaded skills work differently: the full skill content is injected at startup. This contrasts with the standard model where skills are loaded progressively, minimizing the initial context-window footprint.