Run Tests and Fix Failures

Detect test framework, run tests, diagnose and fix failures automatically.

Instructions

Step 1: Detect Test Framework

Check for test configuration files in this order:

File	Framework	Run Command
vitest.config.* or vite.config.* with test	Vitest	npx vitest run
jest.config.* or package.json jest key	Jest	npx jest
phpunit.xml	PHPUnit	./vendor/bin/phpunit
pest.php or Pest in composer.json	Pest	./vendor/bin/pest
pytest.ini / pyproject.toml [tool.pytest]	pytest	pytest
go.mod with _test.go files	Go	go test ./...
Cargo.toml	Rust	cargo test
Gemfile with rspec	RSpec	bundle exec rspec

Also check package.json scripts for test, test:run, test:unit, test:integration.

If CLAUDE.md specifies test commands, use those instead.

Step 2: Run Tests

Run the detected test command with run_in_background=true. Use verbose output flags where available.

Step 3: Analyze Results

If all tests pass, report the summary and stop.

If tests fail, for each failure:

1. Read the failing test to understand what it expects
2. Read the source code being tested
3. Classify the failure:
   • Test is wrong — test expectations don't match intended behavior (update test)
   • Source is wrong — code has a bug (fix source)
   • Both need updating — behavior changed intentionally (update both)
   • Environment issue — missing dependency, stale cache, wrong config (fix setup)

Step 4: Fix and Re-run

For each failure:

1. Make the minimal fix
2. Re-run the specific failing test to verify
3. Move to the next failure

After all individual fixes, run the full suite once more to catch regressions.

Step 5: Report

## Test Results

**Framework:** [detected framework]
**Command:** [command used]
**Result:** X passed, Y fixed, Z remaining

### Fixed
- [test name]: [what was wrong and what was fixed]

### Still Failing (if any)
- [test name]: [diagnosis and why it couldn't be auto-fixed]

Rules

• Never delete or skip a failing test without explicit user approval
• Prefer fixing source over fixing tests (tests define expected behavior)
• If a fix requires changing more than ~20 lines, show the proposed change and ask before applying
• If the same root cause produces multiple failures, fix the root cause once