A Claude-style "skills" system for Mastra agents with progressive disclosure.
npm install mastra-minds
# or
bun add mastra-mindsimport { createMindsAgent, initMindRegistry, FileSystemProvider } from 'mastra-minds';
// Initialize with filesystem provider
await initMindRegistry({
providers: [new FileSystemProvider('./minds')],
});
// Create agent with minds support
const agent = createMindsAgent({
name: 'My Agent',
model: 'gpt-4o',
instructions: 'You are a helpful assistant.',
});
const response = await agent.generate('Help me with...');minds/
└── my-mind/
├── MIND.md # Required: mind definition
├── scripts/ # Optional: executable scripts
└── references/ # Optional: reference docs
MIND.md:
---
name: my-mind
description: Does X when user asks for Y
allowed-tools: Read Write
---
# My Mind
Instructions for the agent...See packages/mastra-minds/README.md for full documentation.
# Install dependencies
bun install
# Run tests
bun test
# Run example
bun examples/quant-agent.tsMIT