Skill Tools

The skill provider exposes a single internal tool, load_skill, that lets an agent pull in the full instructions for one of its attached Agent Skills at runtime. It implements progressive disclosure: the agent first sees only a short <available_skills> list (name + description), then calls load_skill to fetch the complete content for a skill that is relevant to the current task.

Overview

This is not a standalone block. The tool is added to an agent's tool list automatically when one or more skills are attached to an Agent block via its Skills sub-block (a skill-selector). You do not place a skill block on the canvas yourself.

Operations

Configuration

load_skill

No API key, token, or password is required. The tool calls the internal Zelaxy API (POST /api/skills/load) with only a Content-Type: application/json header; authorization comes from the workflow run context, and workspaceId is supplied for you.

Outputs

load_skill

• name (string) — Skill name.
• description (string) — Skill description.
• content (string) — Full skill instructions for the agent to follow.

YAML Example

skill_1:
  type: skill
  name: "Skill"
  inputs:
    operation: "load_skill"
    skill_name: "sql-expert"
  connections:
    outgoing:
      - target: next-block-id

In practice you do not author this block manually. Attach skills to an Agent block's Skills field and the agent decides when to call load_skill itself; the example above just shows the tool's input/output shape. There is no apiKey input — this tool authenticates through the run context, not an environment variable.

Tips

• There is nothing to configure for auth: load_skill hits an internal Zelaxy route, so there is no API key or environment variable to set. The workspaceId is injected automatically — leave it unset.
• Attach skills via the Agent block's Skills sub-block (the skill-selector). Only skills with a non-empty name are passed through to the agent; the agent then sees them in its <available_skills> list and calls load_skill by skill_name when relevant.
• The returned content field holds the full skill instructions. Downstream you can reference the loaded values with {{skill_1.content}}, {{skill_1.name}}, and {{skill_1.description}}, but typically the agent consumes the content internally during the same run rather than passing it on.
• If the named skill cannot be found or loading fails, the tool throws (the underlying /api/skills/load response was not successful), so make sure skill_name exactly matches a name from the available skills list.