Skill Discovery and Progressive Disclosure
A skill becomes useful before its body is loaded. Its name and description earn a place in the catalog; its deeper files earn context only when the task reaches them. Build a filesystem discovery pipeline that separates scope, validation, collision policy, and catalog publication. Explain the three disclosure levels: catalog metadata, active instructions, and task-specific resources. Design references so an agent can reach required detail directly without loading the entire package. Budget catalog space independently from active-skill context. Reject path traversal and symlink escape when a skill reads its own resources. Your agent has 200 installed skills. Loading every SKILL.md, reference file, script, and template at session start would bury the current task in unrelated procedure. Loading nothing would force the user to remember exact filesystem paths. The usual compromise is a catalog: show the model a compact identity and routing description for each eligible skill, then load the full body only after selection. That creates two new engineering problems. First, discovery is not just recursive file search. Skills can exist at project, user, administrator, plugin, or built-in scopes. Two packages can share a name. A symlink can point outside the trusted root. A malformed package can consume catalog space or become impossible to invoke. Second, progressive disclosure can become progressive confusion. If SKILL.md says "read the relevant guide" and…
Skill Discovery and Progressive Disclosure: A skill becomes useful before its body is loaded. Its name and description earn a place in the catalog; its deeper…
This free lesson is part of the AI Engineering from Scratch curriculum. Read the full explanation, run the lesson code, and verify the result in the interactive reader or from the repository source.
Browse the complete course catalog or open this lesson on GitHub.