Skill FAQ
Common questions about Skill authoring, activation, resources, scripts, and evaluation.
Is a Skill just a prompt?
No. A prompt is usually a single instruction. A Skill is a packaged capability: metadata, activation description, workflow instructions, references, scripts, templates, and reusable assets.
How large should SKILL.md be?
Large enough to guide the core task, but small enough to load comfortably. Keep rare background, long examples, and reference material outside the main file.
Why is the description so important?
The agent often sees only the name and description before activation. If the description is vague, the Skill may never load. If it is too broad, it may load for unrelated work.
Should every Skill include scripts?
No. Add scripts only when deterministic automation improves reliability. Good script candidates include parsers, validators, converters, graders, and report generators.
How do Skills relate to MCP?
MCP gives an agent access to tools or data. A Skill tells the agent how to perform a repeatable job. A Skill can instruct the agent to use an MCP-provided capability, but the two solve different layers of the system.
How do I know a Skill actually activated?
Ask for a task that clearly matches the description and watch whether the agent follows the Skill steps rather than improvising. The most reliable check is a task that should not activate: if the Skill fires there too, the description is too broad.
Can two Skills conflict?
They can, and the usual cause is overlapping descriptions. If two Skills both claim to cover "reviewing code", the agent may pick either. Keep each description specific enough that only one Skill can plausibly match a given task.