discovery

Part A: Codebase Scan (EXISTING_PRODUCT mode only)

If mode is EXISTING_PRODUCT, scan the codebase BEFORE interviewing the user. This gives you context to ask smarter questions.

STEP 1: Project Overview Scan

READ (in parallel where possible):
  - README.md or similar documentation
  - package.json / requirements.txt / Cargo.toml / pyproject.toml (dependencies, scripts)
  - CLAUDE.md or project context files
  - Any existing PRD, product docs, or specs in the project
  - .env.example or .env.local (to understand what services are used — DO NOT read .env)

STEP 2: Architecture Scan

SCAN the codebase structure:
  - Use Glob to map the directory structure (src/, app/, pages/, components/, etc.)
  - Identify the tech stack (framework, database, APIs, services)
  - Identify major features by looking at route files, page components, API endpoints
  - Look for authentication, payments, admin panels, or other common product patterns

STEP 3: Feature Inventory

Based on the scan, create a bullet-point inventory:
  - What the product appears to do (core functionality)
  - What tech stack it uses
  - What external services/APIs it integrates with
  - What user-facing features exist
  - What appears to be in progress or incomplete

STEP 4: Present to User

PRESENT the inventory:
  "Based on scanning your codebase, here's what I understand about your product:
  [inventory]

  Is this accurate? What am I missing or getting wrong?"

INCORPORATE corrections before proceeding to the interview.

Part B: Discovery Interview

Walk the user through focused questions. Adapt based on their responses — skip questions they've already answered, dig deeper where they're vague.

Interview Protocol:

• Ask 2-3 questions at a time, not all at once
• Use their language, not PM jargon
• If they give a vague answer, ask a follow-up to sharpen it
• If they say "I don't know," that's fine — log it as an open question
• For existing products, reference what you found in the codebase scan