Skip to main content
This page covers skills for existing hosted SuperAgents during their deprecation. For the official read-only Gigabrain integration in your own AI client, see Gigabrain Skills and Connect Gigabrain.
Skills are how your SuperAgent learns new capabilities. Want it to trade perps on HyperLiquid? Install the hyperliquid skill. Want it to research Polymarket? Install polymarket. Want it to do something no one’s built yet? Write your own in 10 minutes.
SuperAgents use the Agent Skills open standard - the same skill format supported by Claude Code, OpenClaw, Cursor, GitHub Copilot, Windsurf, Gemini CLI, and 26+ other platforms. Any skill built for those platforms works on a SuperAgent, and any skill you write for your SuperAgent works on those platforms.

​
How skills work

A skill is just a folder with a SKILL.md file and optional scripts. The SKILL.md contains markdown instructions that teach the agent what the skill does and how to use it. No SDK, no API wrapper, no boilerplate - just plain language instructions and the tools to back them up.
When your agent encounters a task that matches a skill’s description, it loads the instructions and follows them. Skills are loaded efficiently - metadata first, full instructions only when needed. Inside skill examples, SKILL_DIR means “the directory containing the SKILL.md file being used.” SuperAgents should resolve that path from the opened skill file instead of relying on a pre-injected CLAUDE_SKILL_DIR environment variable.

​
Runtime skills

These ship with every SuperAgent and cover the core trading stack:

Browse SuperAgent runtime skills

Full source code and documentation for every maintained SuperAgent runtime skill.

​
Installing skills

To install a community skill, paste the GitHub link to your agent in chat. The agent will install and sync the skill automatically. Browse skills from any of these registries:

skills.sh

Vercel’s open directory with 110,000+ skills. The largest Agent Skills registry.

ClawHub

OpenClaw’s registry with 2,800+ curated skills.

GitHub

Maintained SuperAgent runtime skills.
Any skill built for Claude Code, OpenClaw, Cursor, or any Agent Skills-compatible platform works on a SuperAgent. Just paste the link.

​
Writing your own skills

You don’t need to be a developer. A skill is just a markdown file that teaches your agent something new. If you can describe it clearly, your agent can learn it.

​
Example: a simple research skill

A skill that gives your agent access to a custom data source. The SKILL.md describes what it does and how to call the script:
The scripts/fear_greed.py file fetches the API and prints the result. Scripts declare their own dependencies via PEP 723 inline metadata and run in isolation via uv run - no dependency conflicts between skills. That’s it. Your agent now knows when to check market sentiment, how to call the script, and how to interpret the results.

​
SKILL.md format

The frontmatter defines metadata: The body is markdown instructions. Keep it under 500 lines - use a references/ directory for detailed documentation that loads on demand.

​
Managing skills

​
Enable or disable

From the agent’s Skills tab on the dashboard, toggle any skill on or off. The agent restarts automatically when skills change. Disabled skills are completely invisible to the agent.

​
Skill tiers

​
Next steps

SuperAgents

Learn about the SuperAgent architecture

Agent Skills Specification

The full open standard specification

SuperAgent Runtime Skills

Browse source code and documentation

ClawHub

Discover community-built skills