Skills to guide Claude Code, Codex, and other coding agents on using the SwanLab AI training experiment tracking platform — both writing tracking code and querying experiment data.
npx skills add SwanHubX/SwanLab-Skill -y -gThen log in to SwanLab:
pip install swanlab
swanlab login # paste your API key from https://swanlab.cn
swanlab ping # (optional) check connectivity
swanlab verify # (optional) validate credentials
npx skillsis a utility for installing skills into major coding agent CLIs. Use--globalto install for all projects, or--agent <name>to target a specific agent. See the npx skills docs for more details.
| Capability | How | Entry point |
|---|---|---|
| Log metrics & rich media (image / audio / text) during training and fine-tuning | Python SDK (swanlab.init / swanlab.log / swanlab.finish + media helpers) |
references/SDK_QUICKSTART.md |
| Inspect a run's metrics, logs, summaries, columns, and media | swanlab api run ... CLI |
references/CLI_REFERENCE.md |
| List / filter experiments by config or summary conditions | swanlab api run filter CLI |
references/CLI_REFERENCE.md |
| Manage projects, self-hosted users, and other resources | swanlab api project / user CLI |
references/CLI_REFERENCE.md |
| Understand the data model & terminology before querying | — | references/SWANLAB_CONCEPTS.md |
| Plot a single experiment's scalar metrics as a line chart | Helper script | scripts/plot_metrics.py user/proj/run -k loss,acc -o chart.png |
| Compare the same metric across multiple experiments (normalization + best-run ranking) | Helper script | scripts/runs_benchmark.py user/proj/r1 user/proj/r2 -k loss --direction lower |
Both helper scripts accept --data file.json to render from previously saved
query output, and require swanlab login (or --api-key / --host).
swanlab.init(mode=...) controls where data goes:
| Mode | Local Storage | Cloud Upload | Use Case |
|---|---|---|---|
online |
Yes | Yes | Normal cloud usage. Requires login. |
local |
Yes | No | Air-gapped / no account needed. |
offline |
Yes | No | Save locally, upload to cloud later via swanlab sync. |
disabled |
No | No | Completely disable all logging. |
Copyright 2024-2025 Emotion Machine (Beijing) Technology Co., Ltd.