Agent Skills
Agent Skills are workspace-scoped instruction sets—markdown bodies with metadata—that agents load when they need a playbook. Unlike Prompts, skills do not take runtime arguments; the body is rendered and injected as-is (with optional Handlebars helpers expanded server-side).
Use skills to codify repeatable workflows: incident response checklists, detection authoring guides, source-specific triage steps, or any instructions you want agents to pull in only when relevant.
In the UI
- In the sidebar, open Agents → Skills.
- Create a skill with a display name, description, and markdown body.
- Toggle Visible to agents when the skill is ready for agents to discover.
- Import a
SKILL.md-style file on the create page to populate metadata and body from YAML frontmatter.
Skills you hide remain editable in the UI but are omitted from agent-facing list and load APIs.
Handlebars helpers
Skill bodies support Handlebars helpers that inline workspace resources at load time:
| Helper | Example | Inlines |
|---|---|---|
query | {{query "failed-logins"}} | Saved query name, description, and SQL |
channel | {{channel "security-alerts"}} | Notification channel metadata |
template | {{template "weekly-summary"}} | Saved template content |
Helpers are resolved when an agent loads the skill, not when you save it in the UI.
MCP and agents
Connected MCP clients can:
- List visible skills (
agent_skills_list) — name and description only - Load a skill by slug (
agent_skills_load) — returns the rendered body
Visible skills also appear as zero-argument prompts in MCP prompts/list, so clients that support MCP prompts can discover them alongside custom prompts.
Scheduled Agents and AI Chat sessions see available skills in their tool context and can call agent_skills_load before following a playbook.
Permissions
| Permission | Access |
|---|---|
| Agent Skills - Read | List skills and view metadata in the UI |
| Agent Skills - Load | Load rendered skill bodies (agents / MCP) |
| Agent Skills - Write | Create, edit, delete, and toggle visibility |
Related
- Prompts — Argumentized MCP prompt templates
- Agents — Scheduled AI automations
- Model Context Protocol — Connect Claude, Cursor, and other MCP clients