Testing Skill

A testing skill for AI coding agents — Claude Code, GitHub Copilot, Cursor, Codex, Gemini CLI, OpenCode, and anything that supports the Agent Skills open standard (SKILL.md).

Write tests at the right level. No bloat, no flaky suites, no misplaced tests. Based on Martin Fowler's testing pyramid — push tests down, justify every E2E test, and delete tests that don't catch real bugs.

What It Does

When an AI agent uses this skill, it will:

• Pick the right test level (unit, integration, E2E) using a decision flowchart
• Reject framework-internals tests, config-value assertions, and duplicate E2E coverage
• Write component tests that assert HTML contracts, not implementation details
• Keep E2E suites focused on user journeys, not feature permutations
• Enforce one happy-path E2E test per journey — edge cases go to integration

Quick Install

One command (GitHub Copilot, Claude Code, and other gh skill agents)

Requires GitHub CLI 2.90.0 or later:

gh skill install nickperkins/testing-skill

This auto-detects your agent and installs to the correct directory. To target a specific agent or scope:

gh skill install nickperkins/testing-skill --agent claude-code --scope user   # personal, all projects
gh skill install nickperkins/testing-skill --agent copilot    --scope project # this repo only

Install script (any agent, no gh required)

curl -fsSL https://raw.githubusercontent.com/nickperkins/testing-skill/main/install.sh | bash

Installs the skill folder to every supported agent directory it finds on your machine. Run --project instead to install into the current repo instead of your home directory:

curl -fsSL https://raw.githubusercontent.com/nickperkins/testing-skill/main/install.sh | bash -s -- --project

Manual install

The skill is a folder containing SKILL.md. Copy the whole folder, not just the file — the folder name is the command name (/testing).

Where to Put It

Each agent looks in a specific directory. The folder name must be testing (it becomes the /testing slash command).

Agent	Personal (all projects)	Project only
Claude Code	~/.claude/skills/testing/SKILL.md	.claude/skills/testing/SKILL.md
GitHub Copilot (VS Code / CLI / cloud agent)	~/.copilot/skills/testing/SKILL.md	.github/skills/testing/SKILL.md
OpenCode	~/.agents/skills/testing/SKILL.md	.agents/skills/testing/SKILL.md
Oh My Pi / Pi Agent	~/.pi/agent/skills/testing/SKILL.md	.pi/skills/testing/SKILL.md

Notes from the official docs:

• Claude Code and Copilot both auto-load skills when relevant, or you invoke with /testing.
• Copilot requires the name field in SKILL.md frontmatter to match the directory name, or the skill silently fails to load.

Cursor

Cursor uses .cursor/rules/*.mdc instead of SKILL.md:

1. Create .cursor/rules/testing.mdc in your project root
2. Paste the SKILL.md body (skip the YAML frontmatter)
3. Set the rule type to Agent Requested

ChatGPT / other tools

1. Copy the SKILL.md body (skip YAML frontmatter)
2. Add to custom instructions or system prompt
3. Ask the AI to "follow the testing skill"

Example Prompts

• "write tests for this feature using the testing skill"
• "review these tests — are they at the right level?"
• "this E2E test is really an integration test — demote it"
• "our E2E suite is too big — which tests should move down?"

The Pyramid in One Picture

          /‾‾‾‾‾‾\
         /  few   \          E2E: very few, real browser/environment
        /    UI    \         Test user journeys that require real UI/runtime
       /‾‾‾‾‾‾‾‾‾‾‾‾\
      /   moderate   \       Integration: moderate count, real deps
     /   integration  \      Test API flows, DB interactions, framework features
    /‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾\
   /     many fast      \    Unit: many, fast, isolated
  /  pure logic & rules  \   Test functions, services, helpers — no external deps
 /________________________\

Default to the lowest layer. The burden of proof is on E2E.

References

• Martin Fowler, "TestPyramid" — https://martinfowler.com/bliki/TestPyramid.html
• Kent Beck, "Test-Driven Development: By Example"
• Google Testing Blog, "Just Say No to More End-to-End Tests"

License

MIT — see LICENSE.