Skills

Skills are reusable, versioned capabilities that agents can invoke. They encapsulate a specific action (e.g. “create-jira-ticket”, “run-security-scan”, “deploy-to-staging”) with a typed input/output schema. Access via Automations → Skills.

What is a skill?

A skill is a self-contained unit of work with:

• Name and slug — human-readable identifier
• Input schema — typed parameters the skill accepts
• Output schema — structured result the skill returns
• Implementation — the logic (tool calls, API requests, code execution)
• Permitted connectors — which integrations the skill can access

Skills are the building blocks used by agents and flows. An agent’s permitted_tools list references skills by ID.

Creating a skill

Via the UI

1. Navigate to Automations → Skills and click + Create.
2. Fill in the builder form: name, description, input/output schema.
3. Define the implementation (tool sequence or code).
4. Validate with a dry-run test input.
5. Publish.

Via API

POST /entities
{
  "kind": "skill",
  "name": "Create Jira Ticket",
  "spec": {
    "slug": "create-jira-ticket",
    "inputs": [
      { "key": "project", "type": "string", "required": true },
      { "key": "summary", "type": "string", "required": true },
      { "key": "priority", "type": "select", "options": ["low","medium","high"] }
    ],
    "outputs": [
      { "key": "ticketKey", "type": "string" }
    ],
    "connectors": ["jira"]
  }
}

Skill kinds

Validation

Before publishing, validate a skill with a sample input:

POST /entities/:id/validate
{ "sampleInput": { "project": "DEMO", "summary": "Test" } }
→ { "ok": true, "details": { ... } }

Versioning

Skills are immutable once published. Each publish creates a new version:

POST /entities/:id/publish
GET /entities/:id/versions

Agents and flows reference a specific version. In-flight runs keep their pinned version.

Linkage

See which agents and flows reference a skill:

GET /entities/:id/linkage
→ { "references": [...], "referencedBy": [...] }

A2A skill scoping

When issuing Agent-to-Agent (A2A) tokens, you scope them to specific skills. A token scoped to skill X cannot invoke skill Y:

POST /a2a/tokens
{ "agentId": "uuid", "skills": ["skill-id"], "ttlSeconds": 3600 }

See also: Agents, Flows, Connectors, Build an MCP