Unit Test Loop

Execute the setup script to initialize the unit test loop:

"${CLAUDE_PLUGIN_ROOT}/scripts/setup-ut-loop.sh" $ARGUMENTS

You are now in a unit test coverage improvement loop.

Please work on the task. When you try to exit, the unit test loop will feed the same PROMPT back to you for the next iteration. You'll see your previous work in files and git history, allowing you to iterate and improve.

Your Task

Each iteration, you must:

1. Run coverage to identify files with low coverage
2. Find ONE important gap - focus on user-facing features, not implementation details
3. Write ONE meaningful test that validates real user behavior
4. Run lint, format, and typecheck the equivalent command in the codebase to ensure code quality
5. Run coverage again to verify improvement
6. Commit with message: test(<file>): <describe behavior>
7. Log progress to .claude/ut-progress.txt

Quality Expectations

Treat ALL code as production code. No shortcuts, no "good enough for now". Every line you write will be maintained, extended, and debugged by others. Fight entropy.

React Testing Library Guidelines

"The more your tests resemble the way your software is used, the more confidence they can give you."

Query Priority (use in order)

1. getByRole - default choice, use name option: getByRole('button', {name: /submit/i})
2. getByLabelText - form fields
3. getByPlaceholderText - only if no label
4. getByText - non-interactive elements
5. getByTestId - last resort only

Query Types

• getBy/getAllBy - element exists (throws if not found)
• queryBy/queryAllBy - only for asserting absence
• findBy/findAllBy - async elements (returns Promise)

Best Practices

• Use screen - screen.getByRole('button') not destructuring render
• Use userEvent.setup() - more realistic than fireEvent
• Use jest-dom matchers - toBeDisabled() not expect(el.disabled).toBe(true)
• Avoid act() - RTL handles it; use findBy or waitFor for async
• Test behavior, not implementation - what users see/do, not internal state

Critical Rules

• ONE test per iteration - focused, reviewable commits
• User-facing behavior only - test what users depend on, not implementation details
• Quality over quantity - a great test catches regressions users would notice
• No coverage gaming - if code isn't worth testing, use /* v8 ignore */ instead
• Log progress - Make sure to log progress to .claude/ut-progress.txt after each successful test run.
• Ensure code quality - Run lint, format, and typecheck before committing

Completion

When the coverage target is reached (or you've covered all meaningful user-facing behavior), output:

<promise>COVERAGE COMPLETE</promise>

IMPORTANT: Only output this promise when it's genuinely true. Do not lie to exit the loop.