ProjectOS wordmark over pink, purple and blue wavy hills

Claude code

Make your AI remember and improve its results

  • Context engineering
  • AI project memory
  • AI workflow
  • Claude Code

What you'll learn

An AI model like Claude works better when the project carries its own rules, memory and QA files.

What you'll get

A folder of .md files, free to download. Drop it into your project and ask your model to implement it.

ProjectOS

A smart workflow for AI models

A set of project files that guides the model through every task, from context and memory to execution and review

ProjectOS flowchart. Claude gets a task, then reads 1. Rules and workflow: Claude.md, the main rules, and Workflow.md, the mandatory process for tasks. Then 2. Project memory: Map.md, where things live; Decisions.md, why it is built this way; History.md, what changed; Backlog.md, what is still open. Claude plans and completes the task, then 3. QA and Review: QA.md, what must be checked; Visual QA.md, live browser checks; Code Review.md, a second look at risky work; Conversations.md, how to report back. Then handoff.

How the files work

Each file has a clear role in how the model thinks, works, and checks its output

1 · Rules and workflow

Scattered paths converging into one straight workflow line ending in a checkmark

Workflow.md

What the file does:

  • The mandatory process for every task.
  • Makes sure the model understands the job before it starts.
  • Keeps each change small, so nothing big breaks by accident.
  • A tiny tweak gets the same care as a whole new page.

2 · Project memory

Site map tree linking one parent page to three child pages

Map.md

What the file does:

  • Draws the whole site on one page. Every screen, every folder.
  • The model reads the file first, so it never guesses where things live.
  • Nothing gets built into the wrong corner of the site.
  • Anyone new finds their way around in minutes.
Decision fork with the chosen path checked and the rejected path faded

Decisions.md

What the file does:

  • Writes down why a choice was made, not only what was done.
  • A new decision openly replaces an old one. Nothing is ever rewritten.
  • The same argument stops coming back every few weeks.
  • Once the user makes a choice, the model remembers it.
Project timeline with milestone dots and a note card recording one change

History.md

What the file does:

  • Keeps a diary of everything that changed, and when.
  • Every job ends with a note on what moved and how to undo it.
  • The model starts every new task knowing what happened before.
  • When something breaks, the diary says where to look.
Loose task notes and a cursor adding them to a backlog tray

Backlog.md

What the file does:

  • Holds everything still open, in one place.
  • Nothing gets added unless you choose to add it.
  • One line per item, in plain words.
  • Finished items are always archived, never erased.

3 · QA & Review

Checklist with three ticked items and a cursor clicking the final checkmark

QA.md

What the file does:

  • Lists what must be checked before a job counts as done.
  • Each kind of change gets its own checklist: text, design, data.
  • The model runs the checks itself, with real clicks in a real browser.
  • Nothing counts as done until it's been tested for real.
Bug under an inspection lens on a screen with faded text lines

Visual QA.md

What the file does:

  • Checks the site by using it in the browser.
  • Every screen, every button, every way out gets pressed.
  • The model searches for invisible bugs.
  • The model behaves like a normal user visiting the site, hunting bugs.
Tilted code page with a bug spotted at its corner during review

Code Review.md

What the file does:

  • Decides how a risky change gets a second look before it ships.
  • Hunts for mistakes that happen in this project, not textbook ones.
  • Turns itself on automatically when a change is big enough.
  • New mistakes get fixed on the spot. Old ones wait for your call.
Speech bubbles pairing a crossed-out reply with a checkmarked reply

Conversations.md

What the file does:

  • Teaches your AI to be easy to read.
  • Model responses turn into short messages, easy to scan fast.
  • Makes responses sound human, not like computer language.
  • Any question the model has for you comes last, never buried.
Free

Get the complete ProjectOS folder

Drop it into your project, ask your model to implement it, and start working with it right away

/ProjectOS

Article by Rotem Elimelech 😎