Agent Skills
sflow ships AI agent skills so coding assistants (Claude Code, Codex, Cursor, GitHub Copilot, and any agent that reads AGENTS.md) can write sflow YAML and debug workflows without you having to explain the schema each time.
The skills are packaged with sflow and exported on demand with the sflow skill command — they are the same content shown on this page, so what an agent runs locally matches what you read here.
If you are an AI agent asked to set up sflow, follow the Setup runbook: it installs sflow and the skills for your harness. Once set up, read the Agent guidelines (AGENTS.md) first, then the relevant skill.
What ships
| Skill | Use it when | Open |
|---|---|---|
| Writing sflow YAML | creating or modifying an sflow.yaml, configuring backends/operators/probes/replicas, setting up inference serving | skill |
| Error analysis | an sflow run fails, you paste an error, or you need to triage logs and task failures | skill |
| AGENTS.md | the step-by-step authoring workflow + hard rules every agent should follow | guidelines |
Each skill also bundles helper scripts (validate_sflow_yaml.py, check_gpu_plan.py, parse_sflow_errors.py, summarize_run.py) that the agent can run. Those scripts are copied into your project by sflow skill; see Setup.
How agents use them
Next
- Set up the skills in your editor / project
- Writing sflow YAML
- Error analysis
- Agent guidelines (AGENTS.md)
- Not sure which feature you need? Open the Feature Map.