---

name: examples-and-recipes description: "Choose, adapt, and troubleshoot Agent Lightning example workflows and optional dependency/backend recipes without reopening original repo examples." disable-model-invocation: true metadata: disco-role: operating license: MIT

Examples and recipes

Use this sub-skill when the user asks which Agent Lightning example fits a task, how to adapt an example, what optional dependencies/backends are needed, or how to add a new example.

Route by task

Request	Read
Pick an example family for a user workflow	references/example-catalog.md
Decide install groups/backends and safe verification level	references/dependency-and-backend-matrix.md
Add or maintain a new example	references/example-contribution-guide.md
Debug optional example failures	references/troubleshooting.md

Key rules

• Most full examples require external services, GPU, datasets, Docker, or long training time. Do not run them by default.
• Start with CPU-local smokes from other sub-skills: authoring, tracing, store, CLI, and endpoint checks.
• Clearly separate "interface can be checked" from "full training semantics verified".
• For optional example workflows, confirm endpoint credentials, model/data availability, hardware, and dependency groups first.
• When creating a new example, include README smoke-test instructions and an "Included Files" section.

Boundary

This sub-skill is a catalog and decision aid. It does not replace focused API guidance in agent-authoring, tracing-and-instrumentation, runner-store-training, or cli-and-services.