Ralph Plan — Create PRD.json

Decompose the user's feature request into a PRD (Product Requirements Document) of atomic, independently implementable stories.

Steps

1. Gather Context

Read the codebase to understand:

• •Project structure and architecture
• •Existing patterns, conventions, test setup
• •CLAUDE.md / project-level instructions
• •Related existing code

2. Clarify Scope

Ask the user:

• •What's the feature? (if not already clear)
• •Any constraints or preferences?
• •Which files/modules are in scope?

3. Decompose into Stories

Break the feature into atomic stories. Each story must be:

• •Independently committable — passes tests on its own
• •Small — one logical change, ideally <200 lines diff
• •Ordered — respects dependencies between stories
• •Minimal dependencies — only add depends_on when there is a genuine code or data dependency (story B modifies files created by story A, or story B's implementation requires story A's output). Stories that touch unrelated files should have depends_on: [] even if they come later in the list. This enables parallel execution.
• •Testable — has clear acceptance criteria

Target: 5-15 stories. If >15, the feature is too large — suggest splitting.

4. Write PRD.json

Write to PRD.json in the current directory:

{
  "feature": "Feature name",
  "description": "One-line description",
  "created": "ISO timestamp",
  "stories": [
    {
      "id": 1,
      "title": "Short imperative title",
      "description": "What to implement",
      "status": "pending",
      "depends_on": [],
      "files": ["path/to/file.py"],
      "acceptance_criteria": [
        "Criterion 1",
        "Criterion 2"
      ],
      "tests": "Description of tests to write or update"
    }
  ]
}

Story Statuses

• •pending — not started
• •in_progress — being worked on
• •done — implemented and committed
• •failed — attempted but failed (needs human review)
• •skipped — intentionally skipped

5. Validate

• •No circular dependencies
• •Every story has at least one acceptance criterion
• •File paths exist (or are clearly new files)
• •Stories are ordered so depends_on references earlier IDs only

6. Report

Print a summary table:

ID | Title                  | Depends On | Files
1  | Add user model         | -          | models/user.py
2  | Add user repository    | 1          | repos/user.py
...

Confirm with user before finalising.