Skills

The Livy skill gives an agent explicit instructions for protected Solana actions.

Use it when the agent supports local skills, instruction packs, or tool-specific runtime guidance.

Public Skill File

The public skill file lives at:

https://livylabs.xyz/SKILL.md

In this repo, the landing site's public static copy belongs in:

apps/landing/public/SKILL.md

What The Skill Enforces

The skill tells the agent to:

• use the Livy MCP endpoint for protected actions
• use a scoped runtime API key
• list available actions before preparing a request
• prepare and authorize before execution
• execute only the approved bundle
• fetch request status after execution
• avoid raw Solana sends for protected workflows

Required Environment

Set these values in the agent runtime:

export LIVY_API_KEY="<runtime-api-key>"
export LIVY_MCP_ENDPOINT="<mcp-endpoint-from-dashboard>"

Do not put admin dashboard credentials or private wallet keys in the agent skill.

Agent Package

When the agent runtime can install packages, use the published @livylabs/ai package:

pnpm add @livylabs/ai

For one-off local checks, use pnpm dlx @livylabs/ai. See MCP for the full livy-ai mcp run flow.

Tool Order

The skill should preserve this order:

livy_list_actions
livy_prepare_action
livy_authorize_action
livy_execute_approved_action
livy_get_request_status

If an agent cannot follow this order, treat the action as outside the protected Livy path.

Read Next

• MCP
• Security & Privacy