Get comfortable with the terminology behind Claude's on-demand agent skills.
0 / 5 completed
1 / 5
At standup, a dev wants Claude to load specialized instructions only when relevant to the task. Which concept fits?
A skill packages instructions, and optionally scripts or resources, that Claude can discover and load on demand for a specific capability. It stays out of the context window until triggered by relevance. This keeps the base prompt lean while extending capability.
2 / 5
During a design review, the team documents when a skill should activate. Where does this live?
A skill's description (metadata) tells the harness when the skill is relevant, similar to a docstring guiding discovery. Claude matches the current task against these descriptions to decide whether to load the skill. Precise descriptions prevent both missed and false-positive activations.
3 / 5
In a code review, a dev bundles a helper script alongside instructions for a skill. What is this pattern called?
Skills can bundle executable resources such as scripts that Claude can run as part of completing the task, not just prose instructions. This lets a skill encapsulate both knowledge and reusable tooling. It distinguishes skills from plain prompt snippets.
4 / 5
An incident report shows two skills with overlapping descriptions both loaded unnecessarily. What is the fix?
Overlapping activation happens when skill descriptions are too broad or similar. Narrowing each description to its specific trigger conditions reduces redundant loading. Clear, non-overlapping descriptions are key to efficient skill discovery.
5 / 5
During a PR review, a teammate distinguishes a skill from a subagent. What is the key difference?
A skill injects relevant instructions into the current conversation's context, while a subagent spins up a separate, isolated context to perform a task and reports back. Skills extend the current agent's knowledge; subagents delegate work entirely. Choosing correctly depends on whether isolation is needed.