Scripts (python3 -m)

One-off utility scripts are run directly as Python modules:

python3 -m unitorch.cli.<module> [args...]

Install UniTorch before running module commands. For local development, use an editable install such as python3 -m pip install -e ., then run commands without PYTHONPATH=src.

---

Copilot Skill Document Exporter

Generates the unitorch-copilot-tools skill package from registered copilot tools into a target folder. These skill documents make unitorch's ML lifecycle capabilities easier for agents to discover, plan, and reuse.

python3 -m unitorch.cli.copilots.skills install all --folder ./skills --force true
python3 -m unitorch.cli.copilots.skills validate --folder ./skills
python3 -m unitorch.cli.copilots.skills uninstall all --folder ./skills

Each tool's output is written to <folder>/unitorch-copilot-tools/<skill-safe-tool-name>/SKILL.md, containing skill frontmatter, CLI/Python usage, parameter metadata, and any remote unitorch-fastapi route metadata declared by the copilot tool. install defaults to ./skills, and uninstall removes the generated copilot skill package from that folder. Use --name all or omit --name to include every registered copilot tool.

The root skills/ layout can then be installed into a local agent skill folder through the external open-agent skills npm package:

npx skills add fuliucansheng/unitorch
npx skills add fuliucansheng/unitorch --folder ./agent-skills

A default external install can report:

{
  "repo": "fuliucansheng/unitorch",
  "folder": "/home/decu/.hermes/skills",
  "copied": [
    "unitorch-config-ini",
    "unitorch-copilot-tools",
    "unitorch-infer-model",
    "unitorch-replace-decorator",
    "unitorch-serve-fastapi",
    "unitorch-train-model"
  ]
}