modal-labs - SKILL.md Agent Skill

name: modal-labs description: Modal Labs (modal.com) — run Python on serverless containers with GPUs, batch jobs, and autoscaling. Precision wrapper to avoid confusion with UI “modal dialogs”. license: Apache-2.0 license metadata: skill-author: Local wrapper (VCO) upstream-skill: modal routing-notes: Prefer this skill over `modal` for auto-routing; reserve `modal` for explicit calls to avoid UI ambiguity.

Modal Labs (modal.com)

Overview

This is a precision wrapper for the upstream modal skill (Modal Labs, modal.com). It exists because “modal” is also a common term for UI dialogs (React/Vue/AntD/etc.).

Use this skill only when the user clearly means Modal Labs (modal.com).

When to Use

Route here when prompts mention one or more of:

modal.com / “Modal Labs”
CLI verbs: modal run, modal deploy, modal serve
serverless containers for Python, batch jobs, autoscaling
GPU workloads (inference, training, rendering) in a serverless setup

Do not use this skill for UI “modal dialog” tasks.

Setup (CLI)

# Install
uv uv pip install modal

# Login (writes token to ~/.modal.toml)
modal token new

Minimal Example

import modal

app = modal.App("hello-modal")

@app.function()
def hello():
    return "hello from Modal"

@app.local_entrypoint()
def main():
    print(hello.remote())

Run:

modal run script.py

Next Actions (choose based on intent)

One-off run: modal run
Long-running endpoint: modal deploy / modal serve
GPU function: add @app.function(gpu="H100") (or another GPU type)

If you need deeper patterns (images, volumes, secrets, web endpoints), follow the upstream modal skill guidance.