feat: Add interactive interview for PRD generation

[Copilot]

Users currently must write PRDs manually, which is counterintuitive given Ralph's autonomous nature. This adds an interview-based workflow where Ralph asks questions and generates PRDs automatically.

Changes

Core Implementation

• ralph-interview.sh: CLI script conducting interactive interviews with colored prompts, input sanitization, and automatic session generation
• /ralph:interview: AI command file for Claude/Cursor with conversation guidelines and example transcripts
• Generates complete sessions: prd.json, progress.txt, learnings.md

Interview Flow

• Session naming → Feature description → Validation commands → Agent/model selection → Task breakdown → Acceptance criteria
• JSON escaping for quotes/backslashes, kebab-case sanitization for session names
• Guides users toward atomic tasks (5-15 min) and measurable criteria

Model Selection

• Clearly asks "Which model should Ralph use?" to specify the model for Ralph's execution
• Shows common model options for Claude (sonnet, opus, haiku) with descriptions
• Explains when to leave blank for agent's default
• Required for cursor agent, optional for others

Documentation

• README/USAGE updated to feature interview as primary session creation method
• docs/INTERVIEW-EXAMPLE.md: Complete walkthrough with example interaction
• Test suite: 11 tests validating script structure, syntax, documentation

Usage

# CLI
./ralph-interview.sh

# AI Command (Claude/Cursor)
/ralph:interview

Ralph interviews user conversationally, then generates:

{
  "branchName": "ralph/user-auth",
  "agent": "claude",
  "model": "sonnet",
  "validationCommands": { "typecheck": "pnpm typecheck" },
  "userStories": [
    {
      "id": "TASK-001",
      "title": "Create auth API routes",
      "acceptanceCriteria": ["Route exists at /api/auth", "Typecheck passes"],
      "priority": 1,
      "passes": false
    }
  ]
}

Makes Ralph self-contained for PRD creation as requested.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>feat: new method to have Ralph ask the user questions instead of only running the loop</issue_title>
<issue_description>Right now, you need to come up with a PRD yourself, which seems counterintuitive, especially when you then run an automated loop for hours using really smart AI agents.

So I've been thinking that it would make a lot more sense to use a method that allows the AI to first interview you about what you want (Ralph-style, using CLIs) and have IT build the PRD for you directly. That would make this package entirely self-contained.

(An even more interesting improvement would be to bake in some git usage to directly clone your project into a worktree or clone overall and have Ralph act on it. That way, you do not even need to keep Ralph in your project directly. Though the sessions should definitely go in there for future compound engineering.)</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes feat: new method to have Ralph ask the user questions instead of only running the loop #21

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[OctavianTocan]

@copilot I want this to also ask user's which model they want to add to the PRD for the Ralph session. By default you can just use Claude Code for the interviewer.

[Copilot]

@copilot I want this to also ask user's which model they want to add to the PRD for the Ralph session. By default you can just use Claude Code for the interviewer.

Done in ad9b302. The interview now explicitly asks "Which model should Ralph use?" with clearer context that this is for Ralph's execution (not the interviewer). It shows common model options (sonnet, opus, haiku) for Claude and explains when to leave blank for the agent's default.