An example spec — a real PRD.md written by nib

An example spec

This is a whole PRD.md, written by nib from someone's notes and shown here exactly as it downloads — nothing tidied for the website. Ten sections, written to be executed rather than read: the non-goals are there so an agent can tell what not to build, the acceptance criteria so it can tell when it is done, and the decisions so it stops reopening them.

1Summary

Build a web-based document reading application for students to access and review research materials remotely. The app will run entirely in a browser, requiring an internet connection, and focus solely on document viewing without editing or creation capabilities. The product will use Next.js for the framework, Cloudflare for hosting, and be developed by a Claude Code agent.

2Problem & target user

The problem is students needing remote access to research documents without downloading them. The target user is students who must review documents online, requiring a reliable, browser-based solution for quick access to academic materials.

3Goals & success metrics

The primary goal is to enable seamless document review without downloads. Success will be measured by users completing 80% of their document reviews without downloading files. No other metrics are specified at this time.

4Non-goals

Do not include offline access. Avoid features for document editing, creation, or collaboration. Do not develop mobile apps or desktop clients. Do not store document content beyond metadata.

5Features

Basic document viewing (PDFs, DOCX, etc.)
Document search within files
User account system for access management
Document metadata storage (e.g., titles, upload dates)

6Tech stack & deployment

Framework: Next.js
Hosting: Cloudflare
Development: Claude Code agent
Application type: Web-only (no mobile/desktop clients)

7Data model

Stores user accounts (basic authentication details)
Stores document metadata (file type, size, upload timestamp, author, title)
No document content storage beyond metadata

8Decisions & constraints

Fixed decisions:
- Web-only platform
Constraints:
- No offline access
- No document editing
Assumptions:
- Target user is students accessing research materials
- Success metric is 80% of reviews completed without downloads
- Document search is required in first version
- User accounts and document metadata are the only data entities
- Web-only platform is the only deployment option

9Phases

Phase 1: Implement basic document viewing and user authentication
Phase 2: Add document search functionality
Phase 3: Optimize performance for large files
Phase 4: Add metadata management and analytics (if needed)

10Instructions for the AI coding agent

Check before starting:
- Confirm all non-goals (offline access, editing, etc.) are excluded
- Verify tech stack matches requirements (Next.js, Cloudflare, Claude Code)
Never change without asking:
- Core focus on document reading (no editing/creation features)
- Data model constraints (no content storage)
Know a change is correct:
- The feature aligns with the first version requirements (basic viewing, search)
- The implementation uses the specified tech stack and avoids non-goals

What €7 changes

Everything above is free and yours to download. The Starter pack sends that spec once through a frontier model and adds the two files your agent expects. Below is section 3 of the very spec you just read, before and after — the real output of a real purchase, not a mock-up.

Section 3, before and after

Freewhat you downloaded

The primary goal is to enable seamless document review without downloads. Success will be measured by users completing 80% of their document reviews without downloading files. No other metrics are specified at this time.

Starter pack€7, one payment

Primary goal: enable a student to complete a document review entirely in the browser, with no file download.

Primary metric (as recorded): 80% of document review sessions are completed without the user downloading the file.

To make that metric verifiable, define the terms:
- Review session: a period beginning when a document is opened in the viewer and ending when the document is closed, the tab is closed, or 30 minutes pass with no page-change, scroll, or search event.
- Completed without download: no `document.download` event was recorded during the session.
- Metric = (sessions with no `document.download` event) / (all sessions of duration >= 10 seconds) over a rolling 7-day window. Sessions under 10 seconds are excluded as accidental opens.
- Target: >= 0.80. Sample: 250 qualifying sessions in a week, 212 with no download event -> 0.848, passing.

Supporting metrics required to interpret the primary one (instrumentation only, no new user-facing features):
- Time to first rendered page, p50 and p95. Budget: p50 <= 2.0 s, p95 <= 5.0 s on a 10 Mbit connection with a 10 MB PDF.
- Search-within-document result latency, p95 <= 1.5 s for a 300-page PDF.
- Viewer error rate (document opened but no page rendered) < 1% of open attempts.

No other metrics are specified. Do not invent engagement or retention goals.

And two files that were not there before

The context filefirst lines

# Project context

A web-based, read-only document viewer for students to review academic PDFs and DOCX files remotely without downloading them, with in-document search and email/password-gated access. Built with Next.js App Router on Cloudflare (Workers/Pages, R2, D1), implemented by a Claude Code agent following the refined PRD as the authority on scope and edge cases.

## Stack

Next.js App Router deployed to Cloudflare Workers/Pages — this is not a Node.js server. No filesystem access, no native Node bindings, no arbitrary npm packages that assume Node APIs; anything needing them must be isolated behind a route explicitly verified to run on the Cloudflare runtime, or dropped. CPU time pe…

tasks.mdfirst lines

# tasks.md

Work top to bottom, respecting `deps`. One task is roughly one session.
Update `status` as you go: todo → doing → done.

---

## T1 — Running Next.js app on Cloudflare with test harness and fixture set
status: todo · deps: —
Scaffold the Next.js App Router project with TypeScript, wire it to deploy on Cloudflare Pages/Workers (via the Cloudflare adapter), set up a test runner (unit + e2e) in CI, and generate/commit the fixture files listed in §10 (300-page text PDF, scanned PDF, password-protected PDF, 0-byte file, truncated file, 150MB/2500-page PDF, 40-page DOCX with tables/images, .pptx, script-tag-titled doc metadata, mixed-page-size PDF).
Done when: `next dev` and a Cloudfla…