Worked Example: Reddit to AI
Reconstructed example — this project shipped before the workflow existed; these docs show what following the workflow produces.
This folder shows what the four-step workflow in this repo actually produces, applied to a real, shipped open-source project: Reddit to AI — a Chrome extension that scrapes Reddit threads, builds a prompt from a preset, previews it with a size meter, and pastes it into Gemini, ChatGPT, Claude, or AI Studio.
Read this first: it's a reconstruction
Reddit to AI shipped before this workflow was written down — nobody produced these documents at the time. What you see here is what the workflow would have produced, rebuilt faithfully from the real project: real feature names, real stack (vanilla JavaScript, Manifest V3, no backend), real layout. Dates line up with the project's actual timeline (first commit: May 2025).
Where the docs would normally cite statistics, they use the workflow's own "Unverified — model knowledge" label instead of inventing sources — exactly what Part 1 tells the AI to do when it has no web access.
Which path it shows
This is the Level A (vibe coder) path — the route most visitors take: plain-English answers, short documents, free tools, and an AI coding assistant doing the heavy lifting. Developer (B) and in-between (C) documents would be longer and more technical, but the shape is identical.
The files, and how they chain
| File | Produced by | What it is |
|---|---|---|
research-reddit-to-ai.md | Part 1 | Deep-research output: concept, users, competitors, budget |
PRD-reddit-to-ai-MVP.md | Part 2 | What the MVP does — and deliberately doesn't do |
TechDesign-reddit-to-ai-MVP.md | Part 3 | How to build it: stack choice with trade-offs, structure, costs |
AGENTS.md | Part 4 (from templates/AGENTS.md) | The filled-in master plan, captured mid-build |
MEMORY.md | Part 4 (from templates/MEMORY.md) | What a living memory file looks like a few weeks in |
Each document ends with a Handoff Context block — the baton in the relay. Part 2 reads it from the research, Part 3 from the PRD, and Part 4 from the tech design, so you never re-answer the same questions. Watch the Source files line grow one arrow per stage, and the Chosen stack / AI coding tool lines appear once the tech design locks them in.
A fun detail: the PRD's Out of Scope list (history, batch mode, multi-language UI, more AI platforms) reads like a spoiler — the real extension shipped most of it later. That's the section working as intended: it's the v2 roadmap in disguise.
Use it like this
- Starting your own project? Compare your Part 1–3 outputs against these to sanity-check depth and tone.
- Notice what "done" looks like: specific features, honest trade-offs, no leftover placeholder brackets, compact pages.
- Don't copy the content — copy the shape.
Share your own
Built something with this workflow? Share it through the Showcase submission issue template (.github/ISSUE_TEMPLATE/showcase_submission.md) — real community examples are how this folder grows.