A skill that acts as a router rather than a single instruction set, dispatching to one of several sub-skills based on the task at hand — the architectural pattern that allows skill libraries to scale past a few dozen entries without becoming unmanageable.
What it is
A flat skill library — create-pdf, parse-pdf, merge-pdf, extract-images-from-pdf — runs into a discovery problem long before it runs into a capability problem. The user has to know which skill applies, how it differs from its siblings, and which combination handles the actual task. After 50 skills, the lookup overhead exceeds the benefit.
Nested skills replace this with a routing layer. A single manage-pdf skill becomes the user-facing entry point. Inside it, sub-skills handle the specifics; the parent skill reads the request and dispatches. Brendan Falk summarises the pattern from community discussion: “instead of separate skills for ‘create PDF’ and ‘parse PDF’, have one skill called ‘manage PDF’ which then routes to the relevant sub-skills.” His claim is that good nesting can scale to 1000+ skills/sub-skills.
Why it matters
Nesting changes what the user has to remember. Without it, every skill is a top-level concept, and the library size is bounded by the user’s working memory. With it, the user only tracks the parent skills (manage PDF, manage code review, manage release); the routing layer handles which specific sub-skill applies. The library grows without the cognitive cost growing.
It also gives the system a place to put cross-cutting logic. A manage-pdf parent can apply consistent input validation, logging, or output formatting across all PDF operations. Without nesting, the same conventions have to be repeated (and risk drifting) across every standalone skill. Nesting and Progressive Disclosure are usually combined: the parent loads lean, and only the relevant sub-skill’s reference files get pulled in for any given task.
Related concepts
- Skills
- Progressive Disclosure
- AI Wiki