What Are Skills?
Skills are reusable blocks of behavior instructions that you attach to agents. They define how an agent should act in specific situations — what process to follow, what rules to enforce, what format to use. Skills are injected into the system prompt at inference time, giving the model explicit guidance without modifying the agent’s core personality.
A single agent can have multiple skills, and a single skill can be shared across many agents. This makes skills the primary mechanism for standardizing behavior across your organization.
Skills vs Context Files
Skills and context files both contribute to the system prompt, but they serve different purposes:
| Skills | Context Files | |
|---|---|---|
| Purpose | Define behavior (what to do) | Provide knowledge (what to know) |
| Content | Instructions, rules, workflows | Facts, specs, documentation |
| Prompt section | [skills] block | [knowledge] block |
| Example | ”When reviewing PRs, check for test coverage" | "Our API uses REST with JSON responses” |
A skill tells the agent how to behave. A context file tells it what’s true about the world.
Creating Skills
In the UI
Settings → Skills → Add Skill — set name, description, scope, and instructions.
Business and Enterprise: for team-scoped skills, assign the skill to the correct team before saving.
Via the admin agent
Create a skill called "Code Review Standards" with scope "org":
When reviewing code:
1. Check for proper error handling — no swallowed errors
2. Verify test coverage for new functions
3. Flag any hardcoded credentials or secrets
4. Ensure consistent naming conventions (camelCase for JS/TS, snake_case for Go)
5. Comment on performance implications for O(n²) or worse algorithms
Skill Scopes
Skills are scoped to control visibility and inheritance:
- Organization (
org) — available to all agents in the organization. Ideal for company-wide standards. - Team (
team) — available to agents used by a specific team. Useful for team-specific workflows. - User (
user) — personal skills that only apply to your own agent sessions. Good for individual preferences.
When multiple scopes apply, they’re all included. An agent might receive org-level coding standards, team-level deployment procedures, and user-level formatting preferences — all in the same system prompt.
Attaching Skills to Agents
Go to Settings → Agents, open the agent, and add skills from the Skills tab. Or ask the admin agent:
Attach the "Code Review Standards" skill to the "Code Reviewer" agent.
An agent can have any number of skills attached.
Managing Skills
Go to Settings → Skills to edit, disable, or delete skills, or ask the admin agent to do the same actions.
Skill Best Practices
- Be specific. “Review code carefully” is too vague. “Check for SQL injection by looking for string concatenation in database queries” is actionable.
- Use numbered steps for procedural skills. Models follow ordered instructions more reliably than unstructured prose.
- Keep skills focused. One skill per concern. A “Code Review” skill and a “Security Audit” skill are better than a monolithic “Code Review + Security” skill.
- Test with real conversations. Create a skill, attach it to an agent, and try edge cases before rolling it out org-wide.