Skip to main content
A Glosso skill is just a directory with TypeScript scripts and a SKILL.md that describes what they do. The agent reads the manifest and knows how to call the scripts.

​
Structure

packages/skills/my-skill/
├── SKILL.md            # machine-readable manifest
├── package.json        # deps for your scripts
└── scripts/
    ├── my-function.ts  # executable script
    └── another.ts

​
1. Write Your Scripts

Each script should be runnable with npx tsx scripts/my-function.ts [args] and print a JSON result to stdout: 
// scripts/my-function.ts
import { GlossoWallet } from '@glosso/core';

const wallet = new GlossoWallet();

// ... do something useful

const result = {
  success: true,
  data: 'whatever the agent needs',
};

console.log(JSON.stringify(result));
Scripts receive arguments from process.argv. Use a simple arg parser or commander: 
import { program } from 'commander';

program
  .argument('<symbol>', 'token symbol')
  .option('--index <n>', 'wallet index', '0')
  .parse();

const [symbol] = program.args;
const { index } = program.opts();

​
2. Write SKILL.md

The SKILL.md frontmatter is the machine-readable part. The body is human + agent documentation. 
---
name: my-skill
description: One sentence describing what this skill does for AI agents.
  IMPORTANT: any critical behavioral instructions for the agent go here.
metadata: {"openclaw": {"emoji": "🔧", "requires": {"bins": ["node"]}}}
---

# SKILL.md — my-skill

## Available Functions

### `my_function(arg1, arg2?)`

**Script:** `tsx scripts/my-function.ts <arg1> [--arg2 value]`

**Purpose:** What this function does and when to use it.

**Parameters:**
| Name | Type | Default | Description |
|---|---|---|---|
| `arg1` | string | — | Required argument |
| `arg2` | string | `default` | Optional argument |

**Returns:**
```json
{
  "success": true,
  "data": "..."
}
When to use: Tell the agent when to call this. 

<Callout type="info">
  The `description` field in the frontmatter is what gets injected into agent prompts. Keep it concise but include any critical behavioral notes (e.g. "NEVER display private keys", "ALWAYS show the explorer link").
</Callout>

---

## 3. Write package.json

```json
{
  "name": "glosso-my-skill",
  "version": "0.1.0",
  "description": "My custom Glosso skill",
  "scripts": {
    "install-deps": "pnpm install"
  },
  "dependencies": {
    "@glosso/core": "workspace:*",
    "@solana/web3.js": "^1.98.0"
  }
}

​
4. Test It

# Run a script directly
npx tsx packages/skills/my-skill/scripts/my-function.ts hello --arg2 world

# Verify it prints clean JSON
npx tsx packages/skills/my-skill/scripts/my-function.ts hello | jq .

​
5. Install Into OpenClaw

Copy your skill to ~/.openclaw/skills/: 
cp -r packages/skills/my-skill ~/.openclaw/skills/my-skill
cd ~/.openclaw/skills/my-skill && pnpm install
Then restart the OpenClaw gateway and your agent can discover it via GLOSSO.md.
Scripts should be idempotent where possible. The agent may retry a tool call if it doesn’t get a response — design scripts so double-execution doesn’t cause unintended effects.