skilld
All skills
nvidia avatar

/earth2studio-create-prognostic

@4cb1092
by NVIDIA Corporationnvidia/skills3.5k stars
424

Create Earth2Studio prognostic (time-stepping forecast) model wrappers. Do NOT use for diagnostic models, data sources, or installation.

Ask your Agent

Use this Skill: https://skilld.dev/gh/nvidia/skills/earth2studio-create-prognostic

This session only. Nothing lands on disk.

referencespr-comment-template.md

≈555 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Reference Comparison Validation

Model: ClassName Comparison date: YYYY-MM-DDTHH:MM:SS Environment: GPU / CPU validation run (do not include machine names, hostnames, absolute paths, or device inventory)

Results Summary

Metric Step 1 (+Xh) Step N (+Yh)
Max absolute difference VALUE VALUE
Mean absolute difference VALUE VALUE
Correlation VALUE VALUE
Finite fraction VALUE VALUE

Key findings:

  • BULLET_SINGLE_STEP_AGREEMENT
  • BULLET_MULTISTEP_OR_AUTOREGRESSIVE_BEHAVIOR
  • BULLET_SPATIAL_PATTERN_QUALITY

Reference Plots

Do not upload or attach images from the automation. Leave placeholders for the PR author to upload images manually in the browser.

Single-step sanity plots:

TODO: Upload SINGLE_STEP_PLOT_NAME_1 here.

TODO: Upload SINGLE_STEP_PLOT_NAME_2 here.

Multi-step vanilla-vs-Earth2Studio comparison plots. For each selected variable, the plot should have forecast lead times as columns and these rows: Earth2Studio wrapper output on top, vanilla reference output in the middle, and relative error on the bottom. Use a small denominator clamp such as max(abs(vanilla), 1e-6).

TODO: Upload REFERENCE_COMPARE_TIMESERIES_VARIABLE_1.png here.

TODO: Upload REFERENCE_COMPARE_TIMESERIES_VARIABLE_2.png here.

TODO: Upload REFERENCE_COMPARE_TIMESERIES_VARIABLE_3.png here.

Validation Scripts

The validation scripts are for review only and must not be committed. Include copy-pasteable Python scripts here, but remove machine names, absolute paths, cache paths, hostnames, and environment-specific details.

<details> <summary>Vanilla reference script</summary>
# Before running this vanilla reference script, clone the original U-CAST
# repository (Rose-STL-Lab/u-cast), then set UCAST_REFERENCE_REPO to that
# checkout or run from a directory containing ./u-cast.
PASTE THE FULL WORKING VANILLA SCRIPT HERE.
</details><details> <summary>Earth2Studio reference script</summary>
PASTE THE FULL WORKING E2S SCRIPT HERE.
</details><details> <summary>Comparison script</summary>
PASTE THE FULL WORKING COMPARISON SCRIPT HERE.
</details>

Source: SKILL.md on GitHub

Third-party checks

2 warnings3mo3 checks · Risk MEDIUM
  • Gen Agent Trust Hub3mo

    This skill facilitates the creation of wrappers for machine learning weather models. It involves fetching and executing third-party inference scripts and uses an unsafe deserialization pattern (torch.load with weights_only=False) which can lead to arbitrary code execution if used with malicious model checkpoints.

  • Socket3mo

    No alerts

  • Snyk3mo

    Risk: MEDIUM · 1 issue

Provenance

Signed by skilld at 4cb1092. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 3 months ago

Capability

version
0.16.0
argument-hint
URL or local path to reference inference script (optional)
Other metadata
metadata
{
  "author": "NVIDIA Earth-2 Team <agent-skills@nvidia.com>",
  "tags": [
    "earth2studio",
    "prognostic-model",
    "python"
  ]
}

README badge

README badge for nvidia/skills/earth2studio-create-prognostic