# sflow

> Declarative workflow descriptor (YAML) with swappable backends. Describe a workflow once
> — tasks, dependencies, resources, and launch methods — and run it locally or on Docker,
> Slurm, or Kubernetes. Install with `uv tool install nv-sflow` (PyPI name `nv-sflow`, command
> `sflow`). sflow ships AI agent skills for writing and debugging sflow YAML, installable
> with `sflow skill`.

## Set up

- [Agent setup runbook (raw markdown)](https://raw.githubusercontent.com/NVIDIA/nv-sflow/main/docs-site/agents-src/setup.md): install sflow, verify it, and install the skills for your agent harness
- [Repository](https://github.com/NVIDIA/nv-sflow)

## Agent skills

- [Agent skills overview](https://nvidia.github.io/nv-sflow/docs/agents/intro): what ships and how agents use it
- [Setup](https://nvidia.github.io/nv-sflow/docs/agents/setup): `sflow skill` install and per-tool wiring
- [Writing sflow YAML](https://nvidia.github.io/nv-sflow/docs/agents/writing-sflow-yaml/): authoring skill
- [Error analysis](https://nvidia.github.io/nv-sflow/docs/agents/sflow-error-analysis/): debugging skill
- [AGENTS.md guidelines](https://nvidia.github.io/nv-sflow/docs/agents/agents-guidelines): authoring workflow and rules

## Docs

- [Why sflow](https://nvidia.github.io/nv-sflow/why-sflow): the workload/backend split, shown on one real example
- [Introduction](https://nvidia.github.io/nv-sflow/docs/user/intro)
- [Quickstart](https://nvidia.github.io/nv-sflow/docs/user/quickstart)
- [Quick reference](https://nvidia.github.io/nv-sflow/docs/user/quick-reference)
- [Configuration](https://nvidia.github.io/nv-sflow/docs/user/configuration)
- [CLI reference](https://nvidia.github.io/nv-sflow/docs/user/cli)
- [Feature map](https://nvidia.github.io/nv-sflow/feature-map)