What does this skill do, and when should you use it?
This Agent Skill guides Earth2Studio installation through uv or pip, optional model extras, environment variables, and verification. It first requires consulting the live installation documentation for the current release tag, extras, and build notes. The agent must never install, upgrade, or modify packages on the user’s behalf. The skill covers installation and environment setup only, not inference code, model selection, workflow composition, or PhysicsNeMo questions.
It asks whether the user prefers uv or pip, whether the package is for a new or existing project, and which Python version is available. It reads the live Earth2Studio installation page for the current version tag, optional extras, and special build requirements. It provides commands for the user to run, helps organize extras by prognostic, diagnostic, data-assimilation, and submodule use, and discusses cache and download-timeout environment variables when relevant. It recommends verifying the installation with import earth2studio and earth2studio.__version__. For installation failures, it directs users to the troubleshooting and FAQ pages and identifies common CUDA, flash-attention, ONNX Runtime, ecCodes, and Python-header issues.
- A developer starting a new Python project can use it to choose between uv and pip for Earth2Studio installation.
- A forecasting user can use it to select model extras for the prognostic models they intend to run.
- A user deploying on a shared filesystem, in CI, or with limited disk space can configure Earth2Studio cache locations and download timeouts.
- A GPU user facing CUDA or native-extension build problems can use its installation-focused troubleshooting guidance.
- A user who has completed installation can use its import and version check to verify the package.
How do you install this skill?
- Commands depend on live NVIDIA documentation and reachable PyPI/GitHub services; direct completion may fail on mainland-China or restricted enterprise networks.
- Add concrete Examples, rollback/uninstall guidance, dependency integrity or source verification, platform compatibility notes, and consistent author/license metadata.
- The BENCHMARK results are reported in the supplied files and do not replace execution or independent reproduction in this static review.
- Shell / CLI
- Network access
- Local filesystem
Python 3.10+uv or pipCUDA-capable GPU with compatible drivers for GPU extras
The skill is located at skills/earth2studio-install/ in the NVIDIA/skills repository. The repository README supports installing one skill with npx skills add nvidia/skills --skill earth2studio-install --yes, or selecting it interactively with npx skills add nvidia/skills. After the skill is installed, the user must run the Earth2Studio uv or pip commands themselves, based on the live installation documentation. The skill forbids the agent from executing package-manager commands.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/earth2studio-install" ~/.claude/skills/
rm -rf "$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:
- Help me install Earth2Studio and choose extras for the models I plan to use.
After placing the skill in an Agent Skills-compatible client, ask something like: “Help me install Earth2Studio and choose extras for the models I plan to use.” The agent should first read https://nvidia.github.io/earth2studio/userguide/about/install.html, then ask about the package manager, project context, and Python version before presenting commands. It should not be used to write inference code, choose models independently, compose workflows, or answer PhysicsNeMo questions.
What are this skill's strengths and limitations?
- Clearly scoped around package installation, optional dependencies, environment configuration, and verification.
- Uses the live official installation documentation as the source for changing version tags, extras, and build instructions.
- Explicitly prevents automatic package installation or environment modification by the agent.
- Covers both uv and pip and calls out several model-specific build and CUDA issues.
- The exact commands and extras list are intentionally dynamic and require access to the live documentation.
- It does not cover inference code, workflow composition, data-source setup, model checkpoint downloads, or unrelated runtime failures.
- Users must execute every installation command themselves and confirm completion.
- The source does not provide a complete operating-system, GPU-model, or model-compatibility matrix.
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 Installation Guide this page ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Earth2Studio Diagnostic Builder ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Earth2Studio Deterministic Forecast ✓ NVIDIA · Official | 53 · 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 |
| Earth2Studio Discover ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
The skill compares uv and pip: uv is recommended and uses a Git source, while pip installs from PyPI and may require manual pre-installation or --no-build-isolation for some extras. The source does not provide a broader comparison with other installation tools.
How did FollowSkills review this skill?
The skill explicitly forbids automatic package installation, upgrades, and modifications, requires the user to run commands, and asks for confirmation. Scope, dependency rationale, and some external data flows are disclosed; no credential theft, covert exfiltration, or malicious defaults are evident. Points are deducted for no rollback plan, dependency-integrity/signature verification, or isolation strategy, and for insufficiently explaining privilege and impact around sudo/conda troubleshooting commands.
The workflow is internally coherent across uv, pip, extras, environment variables, verification, and troubleshooting; BENCHMARK and evals.json document four installation scenarios and expected behaviors. Points are deducted because this review is static, the core workflow depends on undefined WebFetch and live documentation, and there are no local tests, version locking, or sufficiently diagnostic responses for abnormal input.
The audience, installation scenarios, question limit, capability boundaries, and non-fit cases are clearly stated, with distinctions between uv/pip, model extras, and submodules. Points are deducted because Chinese-language interaction guidance is absent, the core flow requires access to PyPI, GitHub, and NVIDIA documentation without mainland-China fallback guidance, and coverage of non-GPU and OS-specific environments is limited.
The frontmatter supplies the name, version 0.16.0, license, author, and tags; the document includes purpose, prerequisites, staged instructions, troubleshooting, and limitations, while skill-card.md supplies references, ownership, and evaluation summaries. Points are deducted because the static validation report identifies a missing Examples section and nonconforming author format; there is no changelog or concrete maintenance/update path, and repository license metadata is NOASSERTION despite other Apache/CC licensing statements.
The skill directly addresses installation, extra selection, environment configuration, and import verification. evals.json defines expected outputs for uv/pip, single-extra, and multi-extra cases, and BENCHMARK reports an overall PASS. Points are deducted because static review cannot confirm current command correctness against live documentation, the user must still choose, execute, and troubleshoot, and no reproducible representative installation result is included; static calibration caps this at 7.
The evidence includes versioned instructions, explicit NVIDIA documentation references, four evaluation tasks, and reported metrics, providing limited traceability. Points are deducted because the benchmark report contains no committed test suite, CI evidence, or independently reproducible artifacts, and live documentation is not pinned with the skill; static calibration caps this at 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 →