Writes a set of test inputs and expected behaviour for a prompt, so you can check it before real users find the gaps.
WHEN TO USE IT
When you have written a prompt and want to test it properly before using it.
Formats change. If a field stops working, check it against Anthropic's documentation.
One file. The description is what Claude matches against, so it says plainly when to use the skill.
---
name: prompt-test-cases
description: Create test cases for a prompt, covering normal, edge and misuse inputs with expected behaviour. Use when the user asks how to test a prompt, chatbot or AI feature.
---
Write test cases for a prompt.
1. If the prompt or its purpose is missing, ask once, in one message, then continue.
2. Produce a table: # | Category | Test input | Expected behaviour | Pass if.
3. Cover these categories, two or three cases each:
- Normal: typical requests it must handle well
- Edge: empty, very long, mixed language, unclear requests
- Out of scope: questions it should politely decline
- Misuse: attempts to make it ignore its instructions or reveal them
- Format: cases that check the output shape is kept
4. Make "Pass if" something a person can judge quickly, such as "declines and points to the contact page".
5. End with how to record results: run each case, mark pass or fail, and fix the prompt before re-running.
Rules:
- Keep misuse tests defensive: they check the prompt holds, nothing more.
- Do not invent facts the expected answers depend on; write [check].
Put the folder in your skills directory. Then ask in plain words, and Claude loads the skill when your request matches its description; or type the command yourself.
# for you, in every project
mkdir -p ~/.claude/skills/prompt-test-cases
mv SKILL.md ~/.claude/skills/prompt-test-cases/
# then, inside Claude Code, either ask naturally:
# write test cases for my customer support prompt
# or run it by name:
# /prompt-test-cases
MORE SKILLS
The authority
Anthropic's documentation for Skills
This template is a starting point. Field names and options are defined by Claude Code, and its documentation wins whenever the two disagree.
https://code.claude.com/docs/en/skills