Claude Code minimal-test-data for compact LLM fixtures
Use minimal representative inputs when exploring extraction, translation, or retrieval, making failures easier to reproduce.
中文Install in Claude Code
Open Refinery or an equivalent test setup in Claude Code. Real model experiments need the project’s configured model environment. The exercise can start with simulated input.
- Review `.claude/skills/minimal-test-data/SKILL.md` and its requirements at the pinned GitHub commit.
- Create `.claude/skills/minimal-test-data/` in your project and save the reviewed file as `.claude/skills/minimal-test-data/SKILL.md`.
- Start Claude Code from that project's root, then select `/minimal-test-data` in the slash menu.
Example task
The small-17 extraction check uses 1 sample article. Its sentence is “Garden walk moved to Friday”. Verify that the extraction retains the date.What to check
Record {processedArticles: 1, expectedDay: 'Friday'} and check the extracted day. Routine CI can reuse the assertion with a mock response.
Original SKILL.md on GitHub · Updated
Frequently asked question
Why start with a small sample instead of a large dataset?
A small fixture that exercises the relevant behavior makes extraction and translation differences easier to debug. Evaluate on larger samples separately.