Skill Format

Oblien skills follow the AgentSkills open standard - the format adopted by Claude Code, Cursor, OpenClaw, Goose, JetBrains AI, and other agent platforms.

Structure

Every skill is a folder containing a SKILL.md file:

oblien-runtime/
└── SKILL.md          # Agent-readable markdown with YAML frontmatter

Skills are single-file by design. The SKILL.md contains everything the agent needs - no supporting code, binaries, or config files.

SKILL.md anatomy

The SKILL.md has two parts: YAML frontmatter and a markdown body.

Frontmatter

---
name: oblien-runtime
description: >-
  Complete Oblien workspace Runtime API - connection and auth, files,
  search, exec, terminal, and watcher.
license: MIT
compatibility:
  - claude-code
  - cursor
  - openclaw
  - goose
  - jetbrains-ai
metadata:
  author: oblien
  version: 1.0.0
  source: https://oblien.com/docs
---

Fields

Field	Required	Description
`name`	Yes	Skill identifier (lowercase, hyphens). 1–64 characters.
`description`	Yes	What the agent learns from this skill. 1–1024 characters.
`license`	No	License identifier (e.g. `MIT`, `Apache-2.0`).
`compatibility`	No	Agent platforms this skill is designed for.
`metadata`	No	Arbitrary key-value pairs - author, version, source URL, etc.

These fields follow the AgentSkills specification. Agents use name and description for discovery (~100 tokens), then load the full body on activation.

Body

The markdown body contains the complete API reference:

Endpoints with URL, method, and parameters
Code examples in SDK, REST, and cURL
Response schemas and field descriptions
Error codes and troubleshooting
Conceptual guides where relevant (networking, auth flows)

The body is plain markdown - no JSX, no framework-specific syntax. Any agent that can read text can consume it.

Progressive disclosure

The AgentSkills standard uses a three-tier loading model:

Tier	What loads	Token cost
Discovery	`name` + `description` from frontmatter	~100 tokens per skill
Activation	Full `SKILL.md` body	Varies (500–5000 tokens recommended)
Resources	Optional `references/` or `assets/` directories	On demand

Oblien skills focus on the first two tiers. The agent sees skill names and descriptions at startup, then loads the full reference when the skill is relevant to the task.

Skill catalog

Skill	Coverage	Recommended for
`oblien-runtime`	Runtime API - files, search, exec, terminal, watcher	Agents working inside a workspace
`oblien-api`	Control-plane API - workspace CRUD, lifecycle, resources	Agents managing workspaces
`oblien-networking`	Networking - firewalls, private links, public access, tokens	Agents configuring network access
`oblien-platform`	Architecture, auth, SDK, images, billing	Agents needing platform context
`oblien-files`	File operations only	Agents focused on file management
`oblien-search`	Search only	Agents focused on code search
`oblien-exec`	Command execution only	Agents running commands
`oblien-terminal`	Terminal sessions only	Agents needing interactive shells
`oblien-watcher`	File watching only	Agents monitoring file changes

See the Skill Catalog for download links and detailed descriptions.

On this page