version: 0.1.0

test-guard

Overview

test-driven-development is the red-green-refactor discipline: write a failing test, write minimum code to pass, refactor. The discipline assumes the test, when red, captures a real behavior gap. When AI writes the test from the same instruction "write tests for X", the discipline breaks:

• The test never goes red (the model writes a test that already passes because the model wrote the implementation logic into the test)
• The test mocks away the very thing it's testing (the model writes a passing assertion against a no-op stub)
• The test tests the model, not the system (the test asserts behavior it would have implemented the same way)

Result: tests are green, the bug ships, "all tests passed" before merge.

test-guard is the post-test-driven-development quality audit. It doesn't replace TDD; it enforces that the tests TDD produced actually test what TDD claims they test.

This skill is the layer-specific complement to ai-code-blindspots (which catches cross-cutting AI blindspots). test-guard is exclusively about test code quality.

When to Use