When you ask your agent to do something that requires specialized knowledge, skills provide that expertise automatically. Here's what happens behind the scenes.
Discovery
Your agent has access to a directory of skills organized by quality and relevance. When you make a request, your agent considers whether any available skills would help.
For common work—building websites, creating videos, analyzing data—your agent recognizes matching skills immediately. For more niche domains, your agent can search the full registry of 950+ skills to find something relevant.
This matching happens automatically. You describe what you want; your agent figures out the best approach, including whether to use a skill.
Activation
When your agent decides a skill would help, it loads the skill's instructions into its working context. These instructions include:
- How to approach the task
- What tools and methods to use
- What the output should look like
- Common pitfalls to avoid
The skill essentially gives your agent a playbook for that specific domain.
Execution
With skill instructions loaded, your agent carries out the work using its normal capabilities—browser sessions, computer sessions, file handling, and so on. The skill guides how these tools are used.
For example, a video animation skill tells your agent:
- Which framework to use (Remotion)
- How to structure the project
- What patterns create good motion graphics
- How to render the final video
Your agent follows this guidance while adapting to your specific request.
Quality Signals
Not all skills are equal. The skill directory includes quality indicators:
- How well-documented the skill is
- How actively it's maintained
- How often it's used successfully
Your agent prefers higher-quality skills when multiple options exist for a task.
Skills and Computer Sessions
Skills are particularly important for computer sessions—the sandboxed environments where your agent runs code and creates files.
When you ask for something like "create a marketing video," the skill tells your agent:
- Which sandbox environment to use (video processing)
- What packages and tools are available
- How to structure the code
- What output formats to produce
- How to verify the work is done correctly
This guidance makes complex work possible. Your agent can produce professional outputs because skills encode best practices for each domain.
When Skills Aren't Used
Not every task needs a skill. Many requests are straightforward enough that your agent handles them with general knowledge:
- Answering questions
- Writing emails
- Simple research
- Basic file operations
- Conversations and planning
Skills come into play when the work requires domain-specific expertise or follows established patterns that are worth encoding.
Limitations
Skills expand what's possible, but they have boundaries:
- Skills can't give agents capabilities they don't have (for example, a skill can't grant access to a platform your agent isn't connected to)
- Very new or obscure domains might not have good skills yet
- Some tasks are too unique to benefit from standardized approaches
When a skill isn't available for something you need, your agent uses general problem-solving. Often that's enough. For recurring specialized needs, new skills can be created—the registry grows continuously.
Related: What Are Skills? | Computer Sessions Overview | Browser Sessions Overview