Ralph Prompt Builder

Ralph Prompt Builder

Build optimized prompts for the ralph-loop plugin with built-in validation and templates.

Quick Start

When the user wants to use ralph-loop, follow this workflow:

1. Determine the Use Case

Ask the user which scenario applies:

• Create new prompt - Guide them through interactive prompt building
• Validate existing prompt - Run validation on their prompt
• Use template - Provide pre-built template for their task type

2. Interactive Prompt Building

If creating a new prompt, ask these questions in sequence:

Question 1: Task Type "What type of task is this?

1. TDD Development (test-driven development workflow)
2. Feature Implementation (new functionality)
3. Bug Fix (fixing existing issues)
4. Greenfield Project (starting from scratch)
5. Other (custom workflow)"

Question 2: Task Description "Describe the task in 1-2 sentences. Be specific about what needs to be accomplished."

Question 3: Completion Criteria "What are the measurable criteria for completion? List specific, verifiable conditions (e.g., 'all tests pass', 'build succeeds', 'coverage >= 80%')."

Question 4: Max Iterations "How many iterations should ralph-loop attempt before stopping? (Recommended: 20-50 for most tasks, 100+ for complex greenfield projects)"

Question 5: Completion Promise "What phrase should Claude output when complete? (Recommended: 'COMPLETE', 'DONE', 'ALL_TESTS_PASS', etc.)"

3. Generate the Prompt

Based on answers, construct a structured prompt following this pattern:

[Task Description]

[Detailed Requirements/Workflow Steps]

Completion Criteria:
- [Criterion 1]
- [Criterion 2]
- [Criterion 3]

When all criteria are met, output <promise>[COMPLETION_PHRASE]</promise>

4. Validate the Prompt

Run scripts/validate-prompt.py to check:

• Completion promise is present
• Criteria are measurable
• Max iterations is set
• No subjective language
• Task doesn't require human judgment

If validation fails, fix issues and re-validate.

5. Present Final Command

Show the user the complete command:

/ralph-loop "[PROMPT_TEXT]" --completion-promise "[PHRASE]" --max-iterations [N]

Template Usage

For common use cases, load the appropriate template from assets/templates/:

• TDD Development → tdd-template.md
• Feature Implementation → feature-dev-template.md
• Bug Fixing → bug-fix-template.md
• Greenfield Projects → greenfield-template.md

Read the template, guide the user through filling placeholders, then validate.

Validation Only

If the user provides an existing prompt to validate:

1. Run scripts/validate-prompt.py with their prompt
2. Present the quality score and findings
3. If issues found, offer to fix them
4. Re-validate after fixes
5. Present the final command

Reference Documentation

For detailed guidance, consult these references as needed:

• ralph-loop-best-practices.md - Comprehensive patterns and anti-patterns for ralph-loop prompts
• completion-criteria.md - Guide for defining measurable, achievable completion criteria
• examples.md - 20+ real-world examples with good vs bad comparisons

When to read references:

• User asks "what makes a good ralph-loop prompt?" → read best-practices.md
• User struggles with completion criteria → read completion-criteria.md
• User wants to see examples → read examples.md
• For complex or unusual tasks → read all three for comprehensive guidance

Multi-Project Context

If the user mentions running multiple projects simultaneously:

1. Ask: "Does this prompt depend on state from other running projects?"
2. If yes: Guide them to reference existing context rather than repeating it
3. Add a "Context" section to the prompt that references shared state
4. Ensure the prompt is self-contained but not redundant

Error Prevention

Never generate a prompt that:

• Lacks a <promise> tag
• Has no max iterations set (unless user explicitly requests unlimited)
• Uses subjective criteria ("make it good", "improve quality")
• Requires human judgment or design decisions
• Cannot be verified automatically

If any of these conditions exist, fix them before presenting the final command.

Workflow Summary

User wants ralph-loop
    ↓
Determine: Create/Validate/Template?
    ↓
[CREATE PATH]              [VALIDATE PATH]           [TEMPLATE PATH]
Ask 5 questions      →     Run validator      →      Load template
Generate prompt            Present findings           Guide fill-in
Validate                   Offer fixes                Validate
Present command            Re-validate                Present command
                          Present command

Important Notes

• Always validate prompts before presenting the final command
• Always set --max-iterations unless user explicitly wants unlimited
• Always use exact <promise>TEXT</promise> format
• Keep prompts focused and measurable
• Prioritize automated verification over manual checks