Skip to content

How to read the reference

This is the document you look at when writing skills in /agent_skills.
The @zio_skill function written here becomes the skillFunction node of the agent builder, and it is called directly by the batch and debugger.

30‑second Quick Start — All tasks follow these four steps

Section titled “30‑second Quick Start — All tasks follow these four steps”
  1. Folder: create /agent_skills/<folder_name>/. One folder is one deployment unit (virtual environment).
  2. Code: inside it write a @zio_skill function in <any_name>.py.
    Data is read and written with Zio.code (basic code) · Zio.node (graph) · Zio.ontology (specification) · Zio.entity (usr table) · Zio.view (virtual view). There is only this SDK for touching data, and SQL·Cypher via Zio.raw() is blocked.
  3. Library (only when needed): list package names in requirements.txt. If the virtual environment already exists and you add more later, run zio deps <folder_name>.
  4. Run — after fixing it, this single line does everything: start and invoke at once.
    Terminal window
    zio run <폴더명>/<함수명> '{"인자":"값"}'
    If nothing appears, use zio logs <folder_name>. The remaining commands (call, restart, ls, ping) are in the “Run inside SSH” section of Create a Skill (@zio_skill).

The places where you write code and where it runs differ. This location has no python or pip; execution is handled by the Skill Runner container via zio run. The .venv/ and main.py are created by the runner, so do not touch them. Engine source or database connection info is also absent here.

If you have no zio, you can perform the same action with curl. The URL is always this one, port 8000:
POST http://api:8000/api/artifacts/skills/call/<folder_name>/execute/<function_name>

Only read the sections you need. Reading everything will make the context feel like a manual. If you’re new, start with Create a Skill (@zio_skill); the rest is for reference.

  • What is this?
  • Zero‑Trust security and isolation design guidelines
  • Zio SDK call API specification
  • Automatic handling of node display names (_display)
  • Main component ecosystem
  • Three forms of a skill
  • Folder structure
  • Using in Agent Builder
  • Intelligent parameter mapping
  • Sample real‑world code writing
  • After building — run it
  • Run inside SSH
  • If you added libraries later
  • What the entry point can do
  • One per entry point
  • Basic code API
  • Create and fix design documents
  • Build according to the design document
  • Ontology graph traversal API
  • Custom table API
  • MongoDB queries
  • What can be queried
  • Fluent CRUD API for collected data
  • Creating tools — external REST API, remote MCP registration
  • Fill in — render
  • Authoring — create · info · update
  • Overview at a glance
  • create — abbreviated spec
  • Order when handling
  • Data collection pipeline
  • $TableName.FieldName
  • #OntologyNodeName.FieldName
  • Appendix 1. How to obtain a Google Spreadsheet service account key
  • There are eight ways to find
  • The label is the entirety of this bundle
  • Overview
  • 🚀 Core features and characteristics
  • Implementation example
  • Overview
  • How to connect and use
  • When multiple people use it together
  • 🪟 Windows user self‑diagnosis
  • Troubleshooting

This folder is read‑only. If you edit it, it will revert on the next deployment.
If something is wrong or missing, tell the engine maintainer.