Project Test Loop

/project:test-loop

When to Use This Skill

Run automated TDD cycle: test → fix → refactor.

Note: Configure project-specific test/build commands in CLAUDE.md or .claude/rules/ for automatic detection.

Steps:

1. Detect test command (if not already configured):

• Check package.json for scripts.test (Node.js)
• Check for pytest or python -m unittest (Python)
• Check for cargo test (Rust)
• Check for go test ./... (Go)
• Check Makefile for test target
• If not found, ask user how to run tests

2. Run test suite:

   # Run detected test command
   [test_command]

3. Analyze results:

If tests FAIL:

• Parse failure output
• Identify failing tests:
• Which tests failed?
• What assertions failed?
• What was expected vs actual?
• Identify root cause:
• Bug in implementation?
• Missing implementation?
• Incorrect test?
• Make minimal fix:
• Fix only what's needed to pass the failing test
• Don't add extra functionality
• Don't fix tests (fix code instead)
• Re-run tests to confirm fix
• Loop back to step 2

If tests PASS:

• Check for refactoring opportunities:
• Code duplication?
• Unclear naming?
• Long functions?
• Complex logic that can be simplified?
• Magic numbers/strings to extract?
• If refactoring identified:
• Refactor while keeping tests green
• Re-run tests after each refactoring
• Ensure tests still pass
• If no refactoring needed:
• Report success
• Stop loop

4. Repeat until:

• All tests pass AND
• No obvious refactoring opportunities

OR stop if:

• User intervention needed
• Blocked by external dependency
• Unclear how to fix failure

5. Report results:

   🧪 Test Loop Results:

   Cycles: [N] iterations

   Fixes Applied:
   - [Fix 1]: [Brief description]
   - [Fix 2]: [Brief description]

   Refactorings Performed:
   - [Refactor 1]: [Brief description]
   - [Refactor 2]: [Brief description]

   Current Status:
   ✅ All tests pass
   ✅ Code refactored
   📝 Ready for commit

   OR

   ⚠️ Blocked: [Reason]
   📝 Next steps: [Recommendation]

TDD Cycle Details:

RED Phase (If starting new feature)

1. Write failing test describing desired behavior 2. Run tests → Should FAIL (expected) 3. This command picks up from here

GREEN Phase (This command handles)

1. Run tests 2. If fail → Make minimal fix 3. Re-run tests → Should PASS 4. Loop until all pass

REFACTOR Phase (This command handles)

1. Tests pass 2. Look for improvements 3. Refactor 4. Re-run tests → Should STILL PASS 5. Loop until no improvements

Common Failure Patterns:

Pattern: Missing Implementation

• Symptom: undefined is not a function, NameError, etc.
• Fix: Implement the missing function/class/method
• Minimal: Just the signature, return dummy value

Pattern: Wrong Return Value

• Symptom: Expected X but got Y
• Fix: Update implementation to return correct value
• Minimal: Don't add extra logic, just fix the return

Pattern: Missing Edge Case

• Symptom: Test fails for specific input
• Fix: Handle the edge case
• Minimal: Add condition for this case only

Pattern: Integration Issue

• Symptom: Test fails when components interact
• Fix: Fix the integration point
• Minimal: Fix just the integration, not entire components

Refactoring Opportunities:

Look for:

• Duplicated code → Extract to function
• Magic numbers → Extract to constants
• Long functions → Break into smaller functions
• Complex conditionals → Extract to well-named functions
• Unclear names → Rename to be descriptive
• Comments explaining code → Refactor code to be self-explanatory

Don't:

• Change behavior
• Add new functionality
• Skip test runs
• Make tests pass by changing tests

Auto-Stop Conditions:

Stop and report if:

• All tests pass + no refactoring needed (SUCCESS)
• Same test fails 3 times in a row (STUCK)
• Error in test command itself (TEST SETUP ISSUE)
• External dependency unavailable (BLOCKED)
• Unclear how to fix (NEEDS USER INPUT)

Loop integrity: this loop's stop condition is the test suite itself — an independent, mechanical judge (a failing test does not care how hard you worked), which is exactly what .claude/rules/loop-integrity.md Pillar 1 asks for. The --max-cycles flag and the "same test fails 3×" rule are the runaway ceiling. Do not make tests pass by editing the tests — that converts the independent judge into a self-judged loop.

Integration with Blueprint Development:

This command applies project-specific skills:

• Testing strategies: Knows how to structure tests
• Implementation guides: Knows how to implement fixes
• Quality standards: Knows what to refactor
• Architecture patterns: Knows where code should go