Skip to main content
Version: v0.2.1

Set up sflow agent skills

The skills are bundled with the sflow package and copied into your project with the sflow skill command, so they always match your installed version.

1. List what's available​

sflow skill --list

Lists the bundled skills (writing-sflow-yaml, sflow-error-analysis) and notes the bundled AGENTS.md workflow guidelines.

2. Copy them into your project​

# Default: copy into ./skills
sflow skill

# Copy into a custom directory
sflow skill -o .cursor/skills

# Overwrite existing bundled skill files when merging
sflow skill -o .cursor/skills --force

sflow skill merges into an existing directory: by default existing files are preserved, and --force overwrites them.

What gets written​

<output-dir>/
  AGENTS.md
  writing-sflow-yaml/
    SKILL.md
    schema-reference.md
    examples.md
    scripts/{validate_sflow_yaml.py, check_gpu_plan.py}
  sflow-error-analysis/
    SKILL.md
    error-catalog.md
    scripts/{parse_sflow_errors.py, summarize_run.py}

3. Wire it into your agent​

Pick the directory your agent already reads:

ToolCommand
Cursorsflow skill -o .cursor/skills
Claude Codesflow skill -o .claude/skills
Any agent using AGENTS.mdsflow skill -o . then keep the generated AGENTS.md at your repo root
tip

AGENTS.md is the entry point: it tells the agent which skill to read for writing YAML vs. debugging, lists the helper scripts, and encodes the hard rules (validate + --dry-run before every run, use file:// artifacts for helper scripts, add failure probes on server tasks, never put credentials in YAML).

4. Keep them current​

Re-run sflow skill -o <dir> --force after upgrading sflow to refresh the bundled skills to the new version.

See also​