Ralph Loop Prompt Builder

Help the user create an effective prompt for /ralph-loop. A good Ralph prompt has three mandatory parts: clear task, verification steps, and completion criteria.

Step 1: Gather Information

Ask the user up to 4 questions using AskUserQuestion. Choose the most relevant questions from this list:

Question 1: Task Type (always ask)

"What kind of task should the Ralph Loop handle?"
Options:

• Write tests
• Fix a bug
• Refactoring
• Implement a new feature

Question 2: Scope (always ask)

"Which files/modules are affected?"
Options:

• Single file (e.g. validators.py)
• One module (e.g. src/services/)
• Multiple modules
• Entire project

Question 3: Verification (ask if not obvious)

"How should Claude verify that progress is correct?"
Options:

• pytest (Unit Tests)
• ruff check (Linting)
• Both (pytest + ruff)
• Other (please describe)

Question 4: Iteration Limit (ask if user didn't specify)

"How many iterations maximum?"
Options:

• 5 (small fix)
• 10 (medium task)
• 20 (larger task)
• 50 (Default, large task)

Step 2: Build the Prompt

Based on the answers, construct a Ralph Loop prompt with this structure:

/ralph-loop max=<N> <TASK_SUMMARY>

Requirements:
1. <Requirement 1>
2. <Requirement 2>
3. <Requirement 3>

Verification after each iteration:
- <verification command 1>
- <verification command 2>

Done when:
- [ ] <Completion criterion 1>
- [ ] <Completion criterion 2>
- [ ] <Completion criterion 3>

Rules for Building the Prompt

Requirements:

• Be specific and measurable, never vague
• Include file paths where relevant
• State what should NOT change (preserve existing behavior)

Verification:

• Must be automated commands Claude can run (pytest, ruff, mypy, etc.)
• Never "check manually" or "make sure it looks good"
• Include both correctness (tests) and quality (linting) checks if applicable

Completion criteria:

• Checkboxes that Claude can verify programmatically
• Each criterion maps to a verification command
• Include "all existing tests still pass" for refactoring tasks

Max Iterations:

• 5 for small, focused fixes
• 10-15 for medium tasks (tests, single-module refactoring)
• 20 for larger tasks (multi-file, new features)
• 50 only for batch operations across many files

Step 3: Present and Refine

Present the generated prompt to the user. Ask if they want to adjust anything:

• Add or remove requirements
• Change iteration limit
• Modify verification steps

Once confirmed, the user can copy-paste the prompt directly.

Examples of Good Output

For "Write tests" + "Single file"

/ralph-loop max=15 Write unit tests for src/services/confidence.py.

Requirements:
1. Every public method of ConfidenceScorer needs at least 2 tests
2. Test edge cases: score 0.0, score 1.0, None values
3. Tests in tests/test_services/test_confidence.py

Verification after each iteration:
- pytest tests/test_services/test_confidence.py -q --tb=short

Done when:
- [ ] At least 10 tests written
- [ ] All tests green
- [ ] Edge cases covered

For "Fix a bug" + "Single file"

/ralph-loop max=8 Fix the bug: <BUG_DESCRIPTION>.

Requirements:
1. Reproduce the bug with a failing test
2. Implement the fix in <FILE>
3. Ensure existing tests still pass

Verification after each iteration:
- pytest <TEST_FILE> -q --tb=short

Done when:
- [ ] Bug-reproducing test exists and is green
- [ ] Existing tests still green
- [ ] No linting errors