What does this skill do, and when should you use it?
This skill guides agents in implementing diagnostic model wrappers for Earth2Studio. It covers simple derived diagnostics, packaged AutoModel diagnostics, and generative or diffusion diagnostics. The workflow includes reviewing reference material, classifying the model, proposing dependencies, writing implementation and tests, and updating project documentation. It is limited to single-time-step transformations and excludes prognostic time-stepping models, data sources, and installation.
Reads a reference inference script, repository, paper, or model documentation and captures inputs, outputs, tensor shapes, normalization, grids, checkpoints, and dependencies; classifies the diagnostic as simple, AutoModel, or generative; creates a model file under earth2studio/models/dx and mock tests under test/models/dx; updates dependency extras and relevant installation, API, and changelog documentation when needed; and runs uv run pytest plus make format, make lint, and make license validation commands.
- An Earth2Studio developer needs to wrap a checkpoint-free derived calculation as a diagnostic model.
- A model developer needs to integrate an AutoModel diagnostic that loads an external checkpoint.
- A researcher needs to add a diffusion, VAE, or stochastic super-resolution diagnostic with sample-count and seeded tests.
- A maintainer needs a repeatable process for coordinate validation, mock testing, optional dependencies, and documentation updates for a new diagnostic model.
How do you install this skill?
- Do not treat the repository benchmark report or target examples as independent execution proof; nothing was executed in this review.
- Before using AutoModel or generative paths, review remote package provenance, commit pinning, licenses, credential permissions, and network reachability, especially from mainland China.
- weights_only=False permits deserialization of full PyTorch objects and should be used only with trusted, verified packages.
- The skill requires approval for dependency-file changes but does not provide equally explicit per-action confirmation or rollback procedures for model downloads, external network access, or GPU use.
- Shell / CLI
- Network access
- Local filesystem
Python 3.10+uvEarth2Studiopytest
Install it from the NVIDIA/skills collection with: npx skills add nvidia/skills --skill earth2studio-create-diagnostic --yes. The README does not document another standalone installation method. This skill is marked Apache-2.0; the repository as a whole is dual-licensed under Apache 2.0 and CC BY 4.0.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/earth2studio-create-diagnostic" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Create an Earth2Studio diagnostic wrapper from this reference script: <URL>
Give the loaded skill a URL or local path to a reference inference script, repository, paper, or model document, for example: "Create an Earth2Studio diagnostic wrapper from this reference script: <URL>". If no reference is supplied, the skill instructs the agent to request one first.
What are this skill's strengths and limitations?
- Covers simple, AutoModel, and generative diagnostic wrapper workflows.
- Specifies coordinate handshakes, mock tests, dependency handling, and focused test commands.
- Provides concrete guidance for single-step execution, generative sampling, and checkpoint loading.
- Covers implementation, tests, documentation, and changelog integration in one workflow.
- Does not support prognostic time-stepping models, data sources, or installation.
- Real package tests may require access to NGC, HuggingFace, S3, or other registries.
- Generative validation may require a GPU and fixed random seeds.
- The supplied material provides no independent test results or broad platform-coverage evidence for this specific skill.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Earth2Studio Diagnostic Builder this page ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Earth2Studio Forecast Wrapper Builder ✓ NVIDIA · Official | 50 · Use with care | ★ 3.8k | today | Apache-2.0 |
| RenderCV Testing Context | 61 · Recommended | ★ 18k | 7mo ago | MIT |
| Earth2Studio Deterministic Forecast ✓ NVIDIA · Official | 53 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Catheter Navigation Fluoroscopy Smoke Tests ✓ NVIDIA · Official | 50 · Use with care | ★ 3.8k | today | Apache-2.0 |
It is distinct from the sibling earth2studio-create-prognostic skill: this skill targets single-step diagnostic transformations, while that alternative targets prognostic time-stepping models.
How did FollowSkills review this skill?
The skill requires reference material, approval before dependency-file edits, credential avoidance, and disclosure of registry, network, and GPU requirements, showing some permission and data-flow awareness. However, external package loading, pickled PyTorch loading with weights_only=False, and remote resources lack complete trust-boundary controls, source verification, confirmation, rollback, and sensitive-data handling, so points are deducted.
The workflow, paths, coordinate conventions, test commands, and exception guidance are fairly structured. However, the referenced supporting files are not present in the supplied evidence, real package tests may depend on network/GPU availability, and complete diagnosable failure behavior or end-to-end reproduction is not statically demonstrated. Under the static cap, the score remains below 10.
The skill clearly separates simple, AutoModel, and generative diagnostics and states coordinate requirements, non-fit cases, and environment prerequisites. Trigger boundaries still depend on user classification, Chinese-language support is not addressed, and core models may depend on NGC, Hugging Face, or S3 services that may be unreliable or unreachable from mainland China, so points are deducted.
It provides a version, Apache-2.0 license, author metadata, layered sections, reference-file index, examples, limitations, troubleshooting, and a self-improvement workflow. Maintenance ownership, formal change history, and update path are not sufficiently explicit, and dependency/reference availability has hidden assumptions, so full marks are not justified.
The workflow covers classification, implementation, testing, registration, documentation, and validation, with concrete examples and deliverables that can support the core wrapping task. However, no independently verifiable representative output is included in the skill itself, and generative/package paths require substantial environment- and model-specific adaptation. The static cap limits this to 7.
The skill states concrete commands, test requirements, coordinate rules, and a prohibition on claiming unrun tests; the supplied benchmark and example code add repository evidence. However, these are primarily self-reported repository materials, without independently reproducible CI, committed key-path test coverage, or third-party corroboration. The static ceiling is therefore 5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →