Skills
Skills are bundles of domain knowledge that teach agents how to approach specific tasks. Unlike tools (which provide actions), skills provide knowledge and context.
The Difference
| Skills | Tools | |
|---|---|---|
| Purpose | Teach how to think | Provide what to do |
| Content | Instructions, best practices | Executable actions |
| Example | “How to debug React apps” | file.write() |
How Skills Work
When you assign a skill to an agent, it learns:
- What to look for — Patterns, signals, indicators
- How to approach problems — Strategies, best practices
- When to escalate — Edge cases, risky situations
- Domain vocabulary — Specific terms and concepts
For example, an “antigravity” skill might teach an agent:
- How to detect IDE agent errors
- Where to look for crash indicators
- Steps to recover from failures
- When to alert the user
Using Skills
Skills are assigned to agents when you configure them. An agent with the right skills can handle domain-specific tasks without detailed instructions each time.
Without skill:
“Check the Antigravity IDE for agent terminated errors. If you find one, look for the retry button and click it. Then wait and verify…”
With skill:
“Monitor Antigravity and auto-recover if needed.”
The skill contains all the detailed knowledge.
Skill Bundles
A skill can include:
- Instructions — What the agent should know
- Specialized tools — Actions specific to this domain
- Examples — Reference cases for the agent
This “atomic” approach keeps detection and recovery bundled together.
Built-in Skills
| Skill | Purpose |
|---|---|
| antigravity | Monitor and heal IDE agent errors |
Creating Custom Skills
Skills are defined as simple markdown files with a YAML header describing metadata. The body contains the instructions that get injected into the agent’s context.
This makes skills easy to write, read, and share.