SKILL (Agent Skill)

Why SKILL Is Needed

Agents can't conjure capabilities out of thin air — they need to acquire "skills" through some means: calling search, executing code, querying databases, sending messages, reading/writing files.

Each type of specific operation corresponds to a Skill. Skill is the definition of what an Agent "can do," exposed to the Agent through the MCP protocol or Tool Calling, enabling it to discover and call these capabilities during reasoning.

---

What Is SKILL

One-line definition: SKILL is a specific capability unit that an Agent can execute, defining "what it can do" and "how to do it."

Analogy: SKILL is like an employee's "job description" — clearly stating what this position does, when to do it, what instructions to accept, and what results to output.

Difference between Skill and Tool:

• Tool is more low-level, referring to a "single operation"
• SKILL is more business-oriented, possibly combining multiple tools

For example, search_web can be a Tool, but a "Intelligence Gathering" Skill might combine search, parsing, and summarization.

---

How to Do It: How to Design Good Skills

Single responsibility: one Skill does one thing, don't stuff too much logic into one skill.

Clear descriptions: the description should let the model accurately determine when to call it, don't just write "fetch data."

Concise parameters: avoid too many required parameters, reducing call failure rates.

Predictable results: unified return value format for easy parsing and processing by the Agent.

Idempotency: same input produces same output for easy retries and debugging.

For Schema writing, see the Function Schema entry — clear Schema is a prerequisite for Skills to be called correctly.

---

Remember this: SKILL is the Agent's "capability list" — clear definitions, explicit descriptions, and single responsibilities enable the Agent to accurately select and call.

Related terms: Agent · Tool Calling