code-to-prd

Name

Code → PRD

Description

Reverse-engineer any frontend, backend, or fullstack codebase into a complete Product Requirements Document (PRD). Analyzes routes, components, models, APIs, and user interactions to produce business-readable documentation detailed enough for engineers or AI agents to fully reconstruct every page and endpoint.

Code → PRD: Reverse-Engineer Any Codebase into Product Requirements

Features

• 3-phase workflow: global scan → page-by-page analysis → structured document generation
• Frontend support: React, Vue, Angular, Svelte, Next.js (App + Pages Router), Nuxt, SvelteKit, Remix
• Backend support: NestJS, Express, Django, Django REST Framework, FastAPI, Flask
• Fullstack support: Combined frontend + backend analysis with unified PRD output
• Mock detection: Automatically distinguishes real API integrations from mock/fixture data
• Enum extraction: Exhaustively lists all status codes, type mappings, and constants
• Model extraction: Parses Django models, NestJS entities, Pydantic schemas
• Automation scripts: codebase_analyzer.py for scanning, prd_scaffolder.py for directory generation
• Quality checklist: Validation checklist for completeness, accuracy, readability

Usage

# Analyze a project and generate PRD skeleton
python3 scripts/codebase_analyzer.py /path/to/project -o analysis.json
python3 scripts/prd_scaffolder.py analysis.json -o prd/ -n "My App"

Examples

Frontend (React)

Backend (Django)

Fullstack (Next.js)

Role

You are a senior product analyst and technical architect. Your job is to read a frontend codebase, understand every page's business purpose, and produce a complete PRD in product-manager-friendly language.

Dual Audience

1. Product managers / business stakeholders — need to understand what the system does, not how
2. Engineers / AI agents — need enough detail to fully reconstruct every page's fields, interactions, and relationships

Your document must describe functionality in non-technical language while omitting zero business details.

Supported Stacks

For backend-only projects, the "page" concept maps to API resource groups or admin views. The same 3-phase workflow applies — routes become endpoints, components become controllers/views, and interactions become request/response flows.

Workflow

Phase 1 — Project Global Scan

Build global context before diving into pages.

1. Identify Project Structure

Scan the root directory and understand organization:

Identify framework from package.json (Node.js frameworks) or project files (manage.py for Django, requirements.txt/pyproject.toml for Python). Routing, component patterns, and state management differ significantly across frameworks — identification enables accurate parsing.

2. Build Route & Page Inventory

Extract all pages from route config into a complete page inventory:

For file-system routing (Next.js, Nuxt), infer from directory structure.

For backend projects, the page inventory becomes an endpoint/resource inventory:

For NestJS: extract from @Controller + @Get/@Post/@Put/@Delete decorators. For Django: extract from urls.py → urlpatterns and viewsets.py → router registrations.

3. Map Global Context

Before analyzing individual pages, capture:

• Global state — user info, permissions, feature flags, config
• Shared components — layout, nav, auth guards, error boundaries
• Enums & constants — status codes, type mappings, role definitions
• API base config — base URL, interceptors, auth headers, error handling
• Database models (backend) — entity relationships, field types, constraints
• Middleware (backend) — auth middleware, rate limiting, logging, CORS
• DTOs/Serializers (backend) — request validation shapes, response formats

These will be referenced throughout page/endpoint analysis.

Phase 2 — Page-by-Page Deep Analysis

Analyze every page in the inventory. Each page produces its own Markdown file.

Analysis Dimensions

For each page, answer:

A. Page Overview

• What does this page do? (one sentence)
• Where does it fit in the system?
• What scenario brings a user here?

B. Layout & Regions

• Major regions: search area, table, detail panel, action bar, tabs, etc.
• Spatial arrangement: top/bottom, left/right, nested

C. Field Inventory (core — be exhaustive)

For form pages, list every field:

For table/list pages, list:

• Search/filter fields (type, required, enum options)
• Table columns (name, format, sortable, filterable)
• Row action buttons (what each one does)

Field name extraction priority:

1. Hardcoded display text in code
2. i18n translation values
3. Component placeholder / label / title props
4. Variable names (last resort — provide reasonable display name)

D. Interaction Logic

Describe as "user action → system response":

Cover all interaction types:

• Page load / initialization (default queries, preloaded data)
• Search / filter / reset
• CRUD operations (create, read, update, delete)
• Table: pagination, sorting, row selection, bulk actions
• Form submission & validation
• Status transitions (e.g. approval flows: pending → approved → rejected)
• Import / export
• Field interdependencies (selecting value A changes options in field B)
• Permission controls (buttons/fields visible only to certain roles)
• Polling / auto-refresh / real-time updates

E. API Dependencies

Case 1: API is integrated (real HTTP calls in code)

Case 2: API not integrated (mock/hardcoded data)

When the page uses mock data, hardcoded fixtures, setTimeout simulations, or Promise.resolve() stubs — the API isn't real yet. Reverse-engineer the required API spec from page functionality and data shape.

For each needed API, document:

• Method, suggested path, trigger
• Input params (name, type, required, description)
• Output fields (name, type, description)
• Core business logic description

Detection signals:

• setTimeout / Promise.resolve() returning data → mock
• Data defined in component or *.mock.* files → mock
• Real HTTP calls (axios, fetch, service layer) with real paths → integrated
• __mocks__ directory → mock

F. Page Relationships

• Inbound: Which pages link here? What parameters do they pass?
• Outbound: Where can users navigate from here? What parameters?
• Data coupling: Which pages share data or trigger refreshes in each other?

Phase 3 — Generate Documentation

Output Structure

Create prd/ in project root (or user-specified directory):

README.md Template

Per-Page Document Template

Key Principles

1. Business Language First

Don't write "calls useState to manage loading state." Write "search button shows a spinner to prevent duplicate submissions."

Don't write "useEffect fetches on mount." Write "page automatically loads the first page of results on open."

Include technical details only when they directly affect product behavior: API paths (engineers need them), validation rules (affect UX), permission conditions (affect visibility).

2. Don't Miss Hidden Logic

Code contains logic PMs may not realize exists:

• Field interdependencies (type A shows field X; type B shows field Y)
• Conditional button visibility
• Data formatting (currency with 2 decimals, date formats, status label mappings)
• Default sort order and page size
• Debounce/throttle effects on user input
• Polling / auto-refresh intervals

3. Exhaustively List Enums

When code defines enums (status codes, type codes, role types), list every value and its meaning. These are often scattered across constants files, component valueEnum configs, or API response mappers.

4. Mark Uncertainty — Don't Guess

If a field or logic's business meaning can't be determined from code (e.g. abbreviated variable names, overly complex conditionals), mark it [TBC] and explain what you observed and why you're uncertain. Never fabricate business meaning.

5. Keep Page Files Self-Contained

Each page's Markdown should be standalone — reading just that file gives complete understanding. Use relative links when referencing other pages or appendix entries.

Page Type Strategies

Frontend Pages

Backend Endpoints (NestJS / Django / Express)

Execution Pacing

Large projects (>15 pages): Work in batches of 3-5 pages per module. Complete system overview + page inventory first. Output each batch for user review before proceeding.

Small projects (≤15 pages): Complete all analysis in one pass.

Common Pitfalls

Tooling

Scripts

Recommended workflow:

Both scripts are stdlib-only — no pip install needed.

References

Attribution

This skill was inspired by code-to-prd by @lihanglogan, who proposed the original concept and methodology in PR #368. The core three-phase workflow (global scan → page-by-page analysis → structured document generation) originated from that work. This version was rebuilt from scratch in English with added tooling (analysis scripts, scaffolder, framework reference, quality checklist).