Earth2Studio Forecast Wrapper Builder
Build and validate Earth2Studio time-stepping forecast wrappers.
The skill requires dependency confirmation, scopes target files, and requires pinned Hugging Face commits; no credential theft, covert exfiltration, or destructive default is evident. Points are deducted because it creates and modifies multiple repository files and runs installation/formatting commands without documenting rollback, data flows, sensitive-data handling, or dependency security review.
The workflow, method order, test commands, and invalid-coordinate tests are fairly explicit. However, the static materials provide no executable reproduction, the BENCHMARK reports an overall FAIL, and method/import conventions may diverge from the target repository. Under static calibration, there is no evidence of complete failure feedback or stable key-path behavior.
The purpose, non-fit cases, optional reference-script input, and expected artifacts are reasonably clear, covering identity and external-model wrapper scenarios. Points are deducted for limited prerequisite, edge-case, and Chinese-language guidance; potential Hugging Face dependence is not accompanied by mainland-China reachability or fallback guidance.
The skill provides a name, version, license, author, tags, staged workflow, examples, troubleshooting, and maintenance-oriented requirements. Points are deducted because the BENCHMARK identifies missing recommended Instructions, prerequisites/requirements, and limitations sections, duplicated reference content, and no clear skill-level changelog, maintenance owner, or update path.
The workflow covers model implementation, tests, dependencies, documentation, validation, and PR preparation, so it could theoretically produce usable wrapper code. Points are deducted because the BENCHMARK verdict is FAIL, Tier 3 results are unavailable, and static material cannot establish that generated code actually passes tests or handles varied model requirements completely.
Fixed revisions, explicit test commands, a static evaluation report, and expected behaviors provide limited auditability. Points are deducted because there is no committed CI/test execution evidence, third-party execution result, or independent reproduction record; the BENCHMARK also lacks Tier 3 signals, so only limited verifiability is supported.
- This is a static-only review; pytest, uv, lint, and model loading were not executed, so the workflow must not be treated as validated.
- Before publication, add explicit prerequisites, limitations, failure recovery/rollback, data-flow disclosure, and alternatives for mainland-China network constraints, and address the failures and duplication recorded in BENCHMARK.
- The skill may read optional external inference scripts and download model packages; review their provenance, commit pinning, dependencies, and network permissions before use.
What does this skill do, and when should you use it?
This skill helps developers connect third-party machine-learning weather models to Earth2Studio's prognostic model interface. It covers reference-script analysis, wrapper implementation, coordinate handling, model loading, testing, documentation, and linting. The workflow requires Earth2Studio's inheritance and coordinate conventions, including an iterator that yields the initial condition first. It is explicitly not intended for diagnostic models, data sources, or installation.
It reads a reference inference script supplied as a URL or local path and analyzes packages, architecture, I/O shapes, time step, resolution, and checkpoint details. It creates earth2studio/models/px/<name>.py with coordinate properties, model loading, forward execution, and a time-stepping iterator. It creates test/models/px/test_<name>.py with mock tests and a real-weight package test. It also guides updates to pyproject.toml, installation documentation, API documentation, CHANGELOG.md, and runs commands such as uv run pytest, make format, and make lint.
- An Earth2Studio developer needs to wrap an existing weather-model inference script as a prognostic model.
- A maintainer needs consistent coordinate validation, lead-time stepping, and iterator behavior for a new forecast model.
- A developer needs mock coverage plus a real-checkpoint forward-pass test for a model wrapper.
- A contributor needs to add the model extra and update installation, API, and changelog documentation.
What are this skill's strengths and limitations?
- Covers the workflow from reference-script analysis through implementation, tests, documentation, and linting.
- Specifies Earth2Studio inheritance, coordinate, lead-time, and iterator conventions.
- Requires both mock tests and a real-weight package test.
- The mandatory uv run convention addresses missing project-dependency issues.
- A reference inference script is required before the core wrapper workflow can begin.
- The scope excludes diagnostic models, data sources, and installation.
- The real-weight test depends on an available checkpoint and network access, while the exact model dependencies remain model-specific.
- Platform details beyond Claude Code and Codex are not documented.
How do you install this skill?
This skill is distributed in the NVIDIA/skills collection. Run npx skills add nvidia/skills and choose the skill and installation destination when prompted. To install it non-interactively by name, run npx skills add nvidia/skills --skill earth2studio-create-prognostic --yes. The source does not document a separate installation command for this individual skill.
How do you use this skill?
Give a compatible coding agent a reference inference-script URL or local path, for example: "Create an Earth2Studio prognostic wrapper from /path/to/inference.py". If no argument is supplied, the workflow asks for the script URL or path. Follow the generated implementation and test workflow, then validate with uv run pytest test/models/px/test_<name>.py -v.