earth2studio-create-diagnostic

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Quick Start Checklist

快速开始检查清单

Do these steps in order. Do not skip ahead. Before editing, read this SKILL.md and load the relevant reference files for the model type. After implementation, run the focused pytest command before saying tests pass. If tests cannot run, report the exact command and failure instead of claiming success.

Read this SKILL.md completely first
Get the reference script, repo, paper, or model documentation (Step 0)
Classify the diagnostic as simple, AutoModel, or generative (Step 1)
Propose dependency extras before editing dependency files (Step 1)
Create
```
earth2studio/models/dx/<name>.py
```
with diagnostic-only APIs
Create
```
test/models/dx/test_<name>.py
```
with mock tests

Run:

uv run pytest test/models/dx/test_<name>.py -m "not package" -v

Add/update model extra, install docs, API docs, and changelog when required

Run:

make format && make lint && make license

Critical command rule: always use

uv run

for Python commands:

Use
```
uv run pytest ...
```
and
```
uv run python ...
```
Do not use bare
```
pytest
```
or
```
python
```
in repo workflows

If the generated model is wrong, do not keep retrying the same fix. Follow Self-Improvement, patch this skill or its references, then continue with the corrected workflow.

请按顺序执行以下步骤，不要跳过。编辑前，请阅读本SKILL.md并加载对应模型类型的相关参考文件。实现完成后，先运行指定的pytest命令，再确认测试是否通过。如果测试无法运行，请报告确切的命令和失败信息，不要声称成功。

先完整阅读本SKILL.md
获取参考脚本、代码库、论文或模型文档（步骤0）
将诊断分类为简单型、AutoModel型或生成型（步骤1）
在编辑依赖文件前，提出依赖扩展建议（步骤1）
创建
```
earth2studio/models/dx/<name>.py
```
文件，仅包含诊断相关API
创建
```
test/models/dx/test_<name>.py
```
文件，编写模拟测试

运行命令：

uv run pytest test/models/dx/test_<name>.py -m "not package" -v

必要时添加/更新模型扩展、安装文档、API文档和变更日志

运行命令：

make format && make lint && make license

关键命令规则：所有Python命令都必须使用

uv run

：

使用
```
uv run pytest ...
```
和
```
uv run python ...
```
在代码库工作流中不要使用裸
```
pytest
```
或
```
python
```
命令

如果生成的模型存在错误，不要重复尝试相同的修复方法。请遵循自我改进部分的指引，修补本技能或其参考文件，然后使用修正后的工作流继续。

Purpose

目的

Implement a diagnostic model wrapper connecting third-party or derived ML transforms to Earth2Studio. Diagnostic models transform data at a single time point: input fields in, output fields out, no forecast integration.

实现诊断模型包装器，将第三方或派生的ML转换与Earth2Studio连接。诊断模型在单个时间点转换数据：输入字段进入，输出字段输出，不涉及预测集成。

Prerequisites

前置条件

Earth2Studio installed via
```
uv
```
with dev dependencies (
```
uv sync --all-extras
```
)
Python 3.10+ environment
Reference inference script, repo, paper, or model documentation
Checkpoint source and license information for packaged models

通过
```
uv
```
安装带有开发依赖的Earth2Studio（
```
uv sync --all-extras
```
）
Python 3.10+环境
参考推理脚本、代码库、论文或模型文档
打包模型的检查点来源和许可证信息

Limitations

限制

Handles single-step transformations only
Does not support time-stepping forecast models; use
```
earth2studio-create-prognostic
```
Real package tests can require network access to NGC, HuggingFace, S3, or other registries
Generative validation can require GPU and fixed seeds for meaningful comparison

仅处理单步转换
不支持时间步长预测模型；请使用
```
earth2studio-create-prognostic
```
实际包测试可能需要访问NGC、HuggingFace、S3或其他注册表的网络权限
生成式验证可能需要GPU和固定随机种子以进行有意义的比较

Diagnostic Model Types

诊断模型类型

Type	Inheritance	Dependency extra	Example
Simple derived diagnostic	`torch.nn.Module` only	Usually none	`Identity` , wind speed
Packaged AutoModel diagnostic	`torch.nn.Module, AutoModelMixin`	Required, even if empty	`PrecipitationAFNO`
Generative diagnostic	`torch.nn.Module, AutoModelMixin`	Required, even if empty	`CorrDiff`

类型	继承关系	依赖扩展	示例
简单派生诊断	仅 `torch.nn.Module`	通常无	`Identity` 、风速
打包的AutoModel诊断	`torch.nn.Module, AutoModelMixin`	必填，即使为空	`PrecipitationAFNO`
生成式诊断	`torch.nn.Module, AutoModelMixin`	必填，即使为空	`CorrDiff`

Workspace

工作区

Context	Location
Harbor eval	Write to `/workspace/output/earth2studio/models/dx/...`
Harbor + `--copy-repo`	Full checkout at `/workspace/repo`
Local clone	Directory with `pyproject.toml`

Never read

evals/targets/

; those files are grader references only.

上下文	位置
Harbor评估	写入至 `/workspace/output/earth2studio/models/dx/...`
Harbor + `--copy-repo`	完整检出至 `/workspace/repo`
本地克隆	包含 `pyproject.toml` 的目录

请勿读取

evals/targets/

；这些文件仅为 grader 参考文件。

Reference Files

参考文件

Load these files on demand during the matching workflow:

File	Content	Load at
`references/skeleton-template.py`	Full diagnostic skeletons for simple, AutoModel, and generative wrappers	Steps 3-6
`references/method-templates.py`	Focused coordinate, loading, forward, and device method snippets	Steps 4-6
`references/testing-guide.py`	Mock, package, exception, sample, and seed test patterns	Step 7
`references/validation-guide.md`	Reference comparison, plots, PR hygiene, and review follow-up	Steps 10-11
`references/pr-body-template.md`	PR body template	Step 11
`references/pr-comment-template.md`	Validation comment template	Step 11

在匹配的工作流中按需加载以下文件：

文件	内容	加载时机
`references/skeleton-template.py`	简单型、AutoModel型和生成型包装器的完整诊断骨架	步骤3-6
`references/method-templates.py`	聚焦坐标、加载、前向传播和设备方法的代码片段	步骤4-6
`references/testing-guide.py`	模拟测试、包测试、异常测试、样本测试和种子测试模式	步骤7
`references/validation-guide.md`	参考对比、绘图、PR规范和评审跟进	步骤10-11
`references/pr-body-template.md`	PR正文模板	步骤11
`references/pr-comment-template.md`	验证评论模板	步骤11

Instructions

操作步骤

Step 0 - Get Reference Material

步骤0 - 获取参考资料

$ARGUMENTS

provides a URL or local path, use it. Otherwise ask:

Please provide a reference inference script, repository, paper, or model documentation.

Capture the reference model's input variables, output variables, tensor shapes, normalization, grid, checkpoint source, dependency requirements, and license.

如果

$ARGUMENTS

提供了URL或本地路径，请使用该路径。否则请询问：

请提供参考推理脚本、代码库、论文或模型文档。

记录参考模型的输入变量、输出变量、张量形状、归一化方式、网格、检查点来源、依赖要求和许可证信息。

Step 1 - Analyze Type and Propose Dependencies

步骤1 - 分析类型并提出依赖建议

Classify the requested diagnostic before editing files:

If the model...	Then use...
Computes a derived quantity with no checkpoint	Simple diagnostic
Loads weights from `Package` or an external checkpoint	AutoModel diagnostic
Produces multiple samples, diffusion outputs, VAE samples, or stochastic super-resolution	Generative diagnostic

Dependency policy:

Simple derived diagnostics usually do not need a
```
pyproject.toml
```
extra.
AutoModel and generative diagnostics must have a named optional dependency extra, even if the list is empty.
Add the extra alphabetically under
```
[project.optional-dependencies]
```
and include it in the
```
all
```
aggregate.

Use the model-extra name in

OptionalDependencyFailure("model-extra")

and

@check_optional_dependencies()

Present the proposed dependency extra and ask the user to approve before editing

pyproject.toml

toml

model-name = ["package1>=version", "package2"]

在编辑文件前，对请求的诊断进行分类：

如果模型...	则使用...
计算派生量且无检查点	简单诊断
从 `Package` 或外部检查点加载权重	AutoModel诊断
生成多个样本、扩散输出、VAE样本或随机超分辨率结果	生成式诊断

依赖策略：

简单派生诊断通常不需要在
```
pyproject.toml
```
中添加扩展。
AutoModel和生成式诊断必须有一个命名的可选依赖扩展，即使列表为空。
在
```
[project.optional-dependencies]
```
下按字母顺序添加扩展，并将其包含在
```
all
```
聚合中。

在

OptionalDependencyFailure("model-extra")

和

@check_optional_dependencies()

中使用模型扩展名称。

在编辑

pyproject.toml

前，先提出拟议的依赖扩展并请求用户批准：

toml

model-name = ["package1>=version", "package2"]

or, when the packaged diagnostic needs no extra runtime packages:

或者，当打包诊断不需要额外的运行时包时：

model-name = []

undefined

model-name = []

undefined

Step 2 - Add Dependencies

步骤2 - 添加依赖

After approval, edit

pyproject.toml

Add the extra alphabetically.
Update the
```
all
```
aggregate.
Prefer minimum supported versions from the reference package documentation.
Do not add broad unpinned Git dependencies unless the reference model requires them and the user approves.

获得批准后，编辑

pyproject.toml

：

按字母顺序添加扩展。
更新
```
all
```
聚合。
优先使用参考包文档中支持的最低版本。
除非参考模型要求且用户批准，否则不要添加宽泛的未固定Git依赖。

Step 3 - Create Model File

步骤3 - 创建模型文件

File:

earth2studio/models/dx/<lowercase>.py

Use the repo-standard SPDX/license header shown in existing model files.

Simple diagnostic imports commonly include:

python

from collections import OrderedDict
import numpy as np
import torch
from earth2studio.models.batch import batch_coords, batch_func
from earth2studio.utils import handshake_coords, handshake_dim
from earth2studio.utils.type import CoordSystem

Packaged and generative diagnostics commonly also include:

python

from earth2studio.models.auto import AutoModelMixin, Package
from earth2studio.models.dx.base import DiagnosticModel
from earth2studio.utils.imports import OptionalDependencyFailure, check_optional_dependencies
from loguru import logger

Canonical method order:

```
__init__
```
```
input_coords
```
```
output_coords
```
decorated with
```
@batch_coords()
```
```
__str__
```
if useful
```
load_default_package
```
for AutoModel/generative diagnostics
```
load_model
```
for AutoModel/generative diagnostics
```
to
```
only when non-PyTorch state must move devices
Private/support methods

__call__

decorated with

@torch.inference_mode()

and

@batch_func()

Avoid shared base classes or broad abstractions unless the wrapper naturally has multiple closely related variants where a small base class reduces duplication.

文件路径：

earth2studio/models/dx/<lowercase>.py

使用现有模型文件中所示的代码库标准SPDX/许可证头部。

简单诊断常用导入：

python

from collections import OrderedDict
import numpy as np
import torch
from earth2studio.models.batch import batch_coords, batch_func
from earth2studio.utils import handshake_coords, handshake_dim
from earth2studio.utils.type import CoordSystem

打包诊断和生成式诊断通常还需导入：

python

from earth2studio.models.auto import AutoModelMixin, Package
from earth2studio.models.dx.base import DiagnosticModel
from earth2studio.utils.imports import OptionalDependencyFailure, check_optional_dependencies
from loguru import logger

标准方法顺序：

```
__init__
```
```
input_coords
```
带有
```
@batch_coords()
```
装饰器的
```
output_coords
```
有用的
```
__str__
```
（可选）
AutoModel/生成式诊断的
```
load_default_package
```
AutoModel/生成式诊断的
```
load_model
```
仅当非PyTorch状态需要移动设备时使用
```
to
```
私有/支持方法

带有

@torch.inference_mode()

和

@batch_func()

装饰器的

__call__

除非包装器自然存在多个密切相关的变体，且小型基类可以减少重复代码，否则避免使用共享基类或宽泛的抽象。

Step 4 - Implement Coordinates

步骤4 - 实现坐标

Diagnostic input coordinates usually use this public Earth2Studio order:

```
batch
```
:
```
np.empty(0)
```
and first in the
```
OrderedDict
```
```
variable
```
: input variable names using Earth2Studio vocabulary names
```
lat
```
: public latitude convention north-to-south, usually
```
90
```
to
```
-90
```
```
lon
```
: public longitude convention
```
0
```
to
```
360
```
, endpoint normally false

No diagnostic wrapper should expose

lead_time

. If a diagnostic needs validity time metadata, document it as per-sample metadata in

coords["time"]

; do not make it a tensor dimension unless an existing dx pattern requires it.

output_coords

must validate inputs with

handshake_dim

and

handshake_coords

. Then update output variables and, when needed, output lat/lon resolution. Generative diagnostics must add a

sample

dimension after

batch

诊断输入坐标通常使用以下Earth2Studio公共顺序：

```
batch
```
：
```
np.empty(0)
```
，且在
```
OrderedDict
```
中排第一
```
variable
```
：使用Earth2Studio词汇命名的输入变量名
```
lat
```
：公共纬度约定（北到南），通常为
```
90
```
到
```
-90
```
```
lon
```
：公共经度约定
```
0
```
到
```
360
```
，通常不包含端点

任何诊断包装器都不应暴露

lead_time

。如果诊断需要有效时间元数据，请将其记录为

coords["time"]

中的每样本元数据；除非现有dx模式要求，否则不要将其作为张量维度。

output_coords

必须使用

handshake_dim

和

handshake_coords

验证输入。然后更新输出变量，必要时更新输出经纬度分辨率。生成式诊断必须在

batch

之后添加

sample

维度。

Step 5 - Implement Forward Pass

步骤5 - 实现前向传播

Use a single-step

__call__

; never create an iterator. Validate coordinates before model execution, then return

(output_tensor, output_coords)

python

@torch.inference_mode()
@batch_func()
def __call__(self, x: torch.Tensor, coords: CoordSystem) -> tuple[torch.Tensor, CoordSystem]:
    output_coords = self.output_coords(coords)
    x = (x - self.center) / self.scale
    out = self.core_model(x)
    return out, output_coords

For generative diagnostics, loop over the batch dimension and generate

number_of_samples

per input item. Use explicit seeds for reproducibility when the reference implementation supports seeded sampling.

使用单步

__call__

；切勿创建迭代器。在模型执行前验证坐标，然后返回

(output_tensor, output_coords)

。

python

@torch.inference_mode()
@batch_func()
def __call__(self, x: torch.Tensor, coords: CoordSystem) -> tuple[torch.Tensor, CoordSystem]:
    output_coords = self.output_coords(coords)
    x = (x - self.center) / self.scale
    out = self.core_model(x)
    return out, output_coords

对于生成式诊断，遍历batch维度，为每个输入项生成

number_of_samples

个样本。当参考实现支持种子采样时，使用显式种子以确保可复现性。

Step 6 - Implement Model Loading

步骤6 - 实现模型加载

For packaged diagnostics:

```
load_default_package
```
should lock HuggingFace URLs to a commit (
```
hf://org/repo@commit
```
) or NGC/S3 versions to an immutable release.
```
load_model
```
should call
```
package.resolve(...)
```
, load checkpoints on CPU first, set modules to
```
eval()
```
, and disable gradients where appropriate.
Use
```
weights_only=False
```
only when loading a pickled full PyTorch object is required.

Decorate optional model classes and

load_model

with

@check_optional_dependencies()

Use
```
loguru.logger
```
for useful loading messages; do not use
```
print()
```
inside
```
earth2studio/
```
.

对于打包诊断：

```
load_default_package
```
应将HuggingFace URL锁定到特定提交（
```
hf://org/repo@commit
```
），或将NGC/S3版本锁定到不可变的发布版本。
```
load_model
```
应调用
```
package.resolve(...)
```
，先在CPU上加载检查点，将模块设置为
```
eval()
```
模式，并在适当的地方禁用梯度。
仅当需要加载完整的PyTorch pickle对象时，才使用
```
weights_only=False
```
。

对可选模型类和

load_model

使用

@check_optional_dependencies()

装饰器。

使用
```
loguru.logger
```
输出有用的加载信息；在
```
earth2studio/
```
内部不要使用
```
print()
```
。

Step 7 - Write Tests

步骤7 - 编写测试

File:

test/models/dx/test_<name>.py

Required tests:

Function	Purpose
`test_<model>_call`	Forward pass with mock or simple model
`test_<model>_exceptions`	Invalid coordinate order, values, or variables raise errors
`test_<model>_package`	Real weights with `@pytest.mark.package` for AutoModel/generative diagnostics

Generative diagnostics also require sample-count and deterministic-seed tests. Use

references/testing-guide.py

. Create a

Phoo<ModelName>

dummy that matches the real core model's interface and produces deterministic output.

Run focused tests:

bash

uv run pytest test/models/dx/test_<name>.py -m "not package" -v
uv run pytest test/models/dx/test_<name>.py::test_<model>_package --package -v

Do not omit package tests for packaged models. If arbitrary random inputs are not physically valid for the real checkpoint, build a stable model-appropriate input while still loading real weights and running a forward pass.

文件路径：

test/models/dx/test_<name>.py

必填测试：

函数	目的
`test_<model>_call`	使用模拟或简单模型进行前向传播测试
`test_<model>_exceptions`	验证无效坐标顺序、值或变量会引发错误
`test_<model>_package`	针对AutoModel/生成式诊断的真实权重测试（带有 `@pytest.mark.package` 标记）

生成式诊断还需要样本数量和确定性种子测试。使用

references/testing-guide.py

。创建一个

Phoo<ModelName>

虚拟类，匹配真实核心模型的接口并生成确定性输出。

运行指定测试：

bash

uv run pytest test/models/dx/test_<name>.py -m "not package" -v
uv run pytest test/models/dx/test_<name>.py::test_<model>_package --package -v

不要省略打包模型的包测试。如果任意随机输入对真实检查点无效，请构建一个稳定的、适合模型的输入，同时仍加载真实权重并运行前向传播。

Step 8 - Register Model

步骤8 - 注册模型

For public models, update

earth2studio/models/dx/__init__.py

alphabetically. Skip registration only when the user explicitly wants an internal or experimental file that should not be exported.

对于公开模型，按字母顺序更新

earth2studio/models/dx/__init__.py

。仅当用户明确希望创建不应导出的内部或实验性文件时，才跳过注册。

Step 9 - Documentation

步骤9 - 文档

For public models:

Add to
```
docs/modules/models_dx.rst
```
alphabetically so API docs include the generated page.

Add to

docs/userguide/about/install.md

if a model extra exists. Include model notes plus both

pip install earth2studio[model-name]

and

uv add earth2studio --extra model-name

instructions.

Update
```
CHANGELOG.md
```
under
```
### Added
```
.

Format and lint:

bash

make format && make lint && make license

对于公开模型：

按字母顺序添加到
```
docs/modules/models_dx.rst
```
，使API文档包含生成的页面。

如果存在模型扩展，请添加到

docs/userguide/about/install.md

。包含模型说明以及

pip install earth2studio[model-name]

和

uv add earth2studio --extra model-name

两种安装指令。

在
```
CHANGELOG.md
```
的
```
### Added
```
部分更新内容。

格式化和检查：

bash

make format && make lint && make license

Step 10 - Validation (if requested)

步骤10 - 验证（如果需要）

references/validation-guide.md

. Create uncommitted vanilla, Earth2Studio, comparison, and sanity-check scripts. Do not commit generated outputs, checkpoints, images, or local validation scripts.

For generative diagnostics, fix seeds and compare matching samples or report statistical/tolerance-based agreement when exact equality is impossible. Ask the user to visually inspect plots before proceeding.

遵循

references/validation-guide.md

。创建未提交的原生脚本、Earth2Studio脚本、对比脚本和 sanity-check 脚本。不要提交生成的输出、检查点、图像或本地验证脚本。

对于生成式诊断，固定种子并对比匹配样本；当无法实现完全相等时，报告基于统计或容差的一致性。请用户在继续前目视检查绘图。

Step 11 - PR (if requested)

步骤11 - 创建PR（如果需要）

references/validation-guide.md

and use:

```
references/pr-body-template.md
```
```
references/pr-comment-template.md
```

Before creating the PR, verify dependency extras,

all

, install docs, API docs, changelog, tests, and validation artifacts are consistent. Do not include machine names, hostnames, absolute paths, cache paths, device inventory, or uploaded image links in PR text. Use plot placeholders for manual image upload.

遵循

references/validation-guide.md

并使用：

```
references/pr-body-template.md
```
```
references/pr-comment-template.md
```

创建PR前，验证依赖扩展、

all

聚合、安装文档、API文档、变更日志、测试和验证工件是否一致。PR文本中不要包含机器名称、主机名、绝对路径、缓存路径、设备清单或上传的图像链接。使用绘图占位符以便手动上传图像。

Examples

示例

Simple Diagnostic

简单诊断

text

User: Create a diagnostic that computes wind speed from u10m and v10m.
Agent: Reads SKILL.md, classifies as simple, creates windspeed.py with only
       torch.nn.Module, writes call and exception tests, runs focused pytest.

text

用户：创建一个从u10m和v10m计算风速的诊断模型。
Agent：阅读SKILL.md，将其分类为简单型，创建windspeed.py文件（仅继承torch.nn.Module），编写调用和异常测试，运行指定的pytest命令。

AutoModel Diagnostic

AutoModel诊断

text

User: Add a precipitation estimator from this reference script.
Agent: Reads SKILL.md and references, proposes dependency extra, creates a
       torch.nn.Module + AutoModelMixin wrapper, writes mock/package tests,
       updates docs/changelog/dependencies, and runs validation commands.

text

用户：根据此参考脚本添加一个降水估算器。
Agent：阅读SKILL.md和参考文件，提出依赖扩展建议，创建继承torch.nn.Module + AutoModelMixin的包装器，编写模拟/包测试，更新文档/变更日志/依赖，运行验证命令。

Generative Diagnostic

生成式诊断

text

User: Wrap this diffusion super-resolution model.
Agent: Classifies as generative, adds sample output coordinates, supports seed
       handling, writes sample and deterministic-seed tests, and prepares seeded
       validation comparisons.

text

用户：包装此扩散超分辨率模型。
Agent：将其分类为生成型，添加样本输出坐标，支持种子处理，编写样本和确定性种子测试，准备种子化的验证对比。

Troubleshooting

故障排除

Error	Solution
`OptionalDependencyFailure`	Install with `uv sync --extra <model-extra>` or fix the extra name
Coordinate handshake fails	Check `OrderedDict` order and `handshake_dim` indices
Wrong output shape	Verify `output_coords` lengths match returned tensor shape
`ModuleNotFoundError: pytest`	Use `uv run pytest` , not bare `pytest`
Package test fails on random input	Use a stable physically plausible input while still loading real weights

错误	解决方案
`OptionalDependencyFailure`	使用 `uv sync --extra <model-extra>` 安装，或修正扩展名称
坐标握手失败	检查 `OrderedDict` 顺序和 `handshake_dim` 索引
输出形状错误	验证 `output_coords` 长度与返回张量形状匹配
`ModuleNotFoundError: pytest`	使用 `uv run pytest` ，不要使用裸 `pytest`
包测试在随机输入上失败	使用稳定的、符合物理规律的输入，同时仍加载真实权重

Reminders

注意事项

Do:

Use
```
uv run python
```
and
```
uv run pytest
```
for all Python commands.
Use
```
@batch_coords()
```
on
```
output_coords
```
.

Use

@torch.inference_mode()

and

@batch_func()

__call__

Keep
```
batch
```
as the first coordinate with
```
np.empty(0)
```
in
```
input_coords
```
.
Validate coordinates with
```
handshake_dim()
```
and
```
handshake_coords()
```
.
Add
```
sample
```
in generative
```
output_coords
```
.
Include the repo-standard SPDX/license header in every Python file.
Use
```
loguru.logger
```
, never
```
print()
```
, inside
```
earth2studio/
```
.

Do not:

Inherit from
```
PrognosticMixin
```
.
Include
```
lead_time
```
coordinates.
Create
```
create_iterator
```
.
Create general base classes for a single wrapper without a clear multi-variant need.
Commit API keys, credentials, validation scripts, plots, or generated outputs.
Read from
```
evals/targets/
```
.

请遵循：

所有Python命令使用
```
uv run python
```
和
```
uv run pytest
```
。
在
```
output_coords
```
上使用
```
@batch_coords()
```
装饰器。

在

__call__

上使用

@torch.inference_mode()

和

@batch_func()

装饰器。

在
```
input_coords
```
中保持
```
batch
```
为第一个坐标，值为
```
np.empty(0)
```
。
使用
```
handshake_dim()
```
和
```
handshake_coords()
```
验证坐标。
在生成式
```
output_coords
```
中添加
```
sample
```
维度。
在每个Python文件中包含代码库标准的SPDX/许可证头部。
在
```
earth2studio/
```
内部使用
```
loguru.logger
```
，切勿使用
```
print()
```
。

请勿：

继承自
```
PrognosticMixin
```
。
包含
```
lead_time
```
坐标。
创建
```
create_iterator
```
。
为单个包装器创建通用基类，除非明确存在多变体需求。
提交API密钥、凭证、验证脚本、绘图或生成的输出。
读取
```
evals/targets/
```
目录下的文件。

Self-Improvement

自我改进

If this skill produces incorrect outputs, update it before continuing:

Identify the issue in the generated code or workflow.
Edit
```
SKILL.md
```
or the relevant file in
```
references/
```
to fix the guidance.
Run focused validation for the changed skill files.
Commit the skill fix separately when working in a branch that expects commits.
Continue the model implementation with the corrected workflow.

如果本技能生成的输出不正确，请先更新技能再继续：

识别生成代码或工作流中的问题。
编辑
```
SKILL.md
```
或
```
references/
```
中的相关文件以修正指引。
对修改后的技能文件运行指定的验证。
在需要提交的分支中，单独提交技能修复。
使用修正后的工作流继续模型实现。