使用方法
- 复制安装命令，在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物，并对照来源检查输出。
npx skills add https://github.com/justcatit/cat-skills --skill cat-smoke-tests开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况，不等于独立质量或安全测试。
按源仓库内容呈现，保留标题、案例、代码、表格、链接以及原文引用的演示图片。
Smoke tests
Is there data? One row is enough, the test costs almost nothing, and when it fails nothing else is worth running. Every fact is on the documentation page it links; the same page as markdown is at the URL with index.md appended, which is the form to read when your fetch tool summarizes HTML.
The pattern
Three grains: the table, the latest partition or period, a business slice. The middle one catches the load that did not run; it runs first after every load and before the expensive comparisons. The tests, the grains, and the pairing with a freshness check when "not empty" is not enough: https://docs.justcat.it/how-to-guides/test-patterns/smoke-tests-with-set-is-not-empty/
The expectation and the one trap it has: https://docs.justcat.it/reference/tests/expectations/set-is-not-empty/
Every table, without writing every test
One template over the catalog (the tables of a schema, the partitions of a control table, the customers), tagged smoke, and the table created next week gets its test on the next open; load the skill for generating tests from metadata: https://docs.justcat.it/reference/tests/templates/
Run the smoke suite alone with the tag filter before anything else: https://docs.justcat.it/how-to-guides/organize-and-run-tests/run-your-tests/
The traps
SELECT COUNT(*)withset is not emptyalways passes: one row always comes back.
Select a row (TOP 1, LIMIT 1), not a count.
- A table with one stale row passes; pair the smoke test with a freshness rule (`set is
empty on the newest timestamp being too old) or an exact set row count` where the count is known.
- A smoke test per table that does not use a template is a test the new table never gets.
