---
title: "Comment by Abraam Gotsiridze on \"Vibe Coding a Complete MERN App: A Phase-by-Phase Playbook\" | Hashnode"
description: "The line I'd pin above the desk is in your iteration rules: update the docs when a decision changes, or the agent will quietly revert your new decision to the o"
url: "https://hashnode.com/posts/vibe-coding-a-complete-mern-app-a-phase-by-phase-playbook/6ae48e8d39ff6383e1a3cbd1/comment/6aeca4351e4fa3cde7c8ac0a"
---

[Discussion](https://hashnode.com/posts/vibe-coding-a-complete-mern-app-a-phase-by-phase-playbook/6ae48e8d39ff6383e1a3cbd1)

Comment [Vibe Coding a Complete MERN App: A Phase-by-Phase Playbook](https://hashnode.com/posts/vibe-coding-a-complete-mern-app-a-phase-by-phase-playbook/6ae48e8d39ff6383e1a3cbd1)

[Abraam Gotsiridze](https://hashnode.com/@abraamgotsiridze)

[Nov 5](https://hashnode.com/posts/vibe-coding-a-complete-mern-app-a-phase-by-phase-playbook/6ae48e8d39ff6383e1a3cbd1/comment/6aeca4351e4fa3cde7c8ac0a)

The line I'd pin above the desk is in your iteration rules: update the docs when a decision changes, or the agent will quietly revert your new decision to the old one. That one often gets blamed on the model, when really the context files are still describing last month's project.

I also like how "decisions, not adjectives" for context.md pairs with "checkable in a diff" for AGENTS.md. "Always scope task queries by req.userId" is something a reviewer can verify. "Write secure code" isn't.

One small addition for the feature prompt: when the spec changes, ask the agent to list which lines in context.md and AGENTS.md the change makes untrue, not only to add the new user story. The filter tabs are purely additive, but a change like moving the token to an httpOnly cookie would leave the auth section of context.md still saying localStorage, and the next fresh session would follow it.

Do you ever go back to idea.md once the PRD exists, or is it purely there for history?

[Asfaq Leeon](https://hashnode.com/@leeonscoding)

[Nov 8](https://hashnode.com/posts/vibe-coding-a-complete-mern-app-a-phase-by-phase-playbook/6ae48e8d39ff6383e1a3cbd1/comment/6aeca4351e4fa3cde7c8ac0a#reply-6af0b2089fa854247f1022ee)

Thanks, this is a sharp read. Stale context files really are the most common thing blamed on the model. Your addition is a good one. I only covered the additive case, and the dangerous changes are the ones that invalidate existing text. The httpOnly cookie example is exactly it: the auth section in context.md keeps saying localStorage, and the next fresh session faithfully builds on the old decision. I'll add a line to the feature prompt along the lines of: 'before implementing, list every line in context.md, AGENTS.md and the PRD that this change makes untrue, update them, then write the code.' The idea.md basically keeps the product's idea/vision in a short form. So, it's a history. You might change your requirements several times. You may ask AI to add features, remove functionality, change the architecture, or redesign the UI. Eventually, the original vision can get lost. The idea.md helps prevent that.
