Skip to content
ONECLICKTOOLINDEPENDENT STUDIO

How Did I Run This Project Again? Write the Command Down

npm run dev? make serve? Which env file? You forget how to start your own projects. Why it happens with many repos and a tiny habit that ends it.

By · Oct 18, 2026 · 2 min read

RedditXLinkedInFacebook
A developer with glasses holds his head while reading an old printed note, trying to recall how to start a project.
On this page · 7 sections
  1. The short answer
  2. Why you forget your own commands
  3. Write a "how to run" block
  4. Keep it where you look first
  5. Let the agent keep it current
  6. Keep reading
  7. Frequently asked questions

You come back to a project to fix one small bug. npm start? Missing script. npm run dev? It starts, then fails — something about an env variable. Was there a .env.example? Did you need the backend running first? Which port?

Twenty minutes later it's running, and you've forgotten what the bug was.

Part 6 of No Process — the smallest missing document in most side projects.

The short answer#

  • With many repos, you forget how to start each one: the command, the env file, the port, the one step that must run first.
  • Write a 'how to run' block in every repo — the exact commands, copy-pasteable, in order.
  • Keep it next to the project's state, where both you and the agent look first.
  • Ask the agent to update it whenever setup changes, and test it on a fresh clone now and then.

Why you forget your own commands#

With one project you run every day, the commands live in your fingers. With many projects, each on its own stack, touched every few weeks, they don't:

  • The app uses npm run dev, the game uses a Unity build, the book uses a Python script.
  • One needs an env file, one needs a simulator booted first, one needs a specific Node version.
  • The agent set half of it up for you, so you never learned it in the first place.

Every return starts with rediscovery — the same cost as coming back after a break, but for setup.

Write a "how to run" block#

In every repo, write the exact commands, in order, copy-pasteable:

How to run
  cp .env.example .env        # fill API_URL
  npm install
  npm run dev                 # http://localhost:4321
Tests
  npm test

No prose, no explanations unless something is genuinely surprising. The goal is that you — or an agent — can paste it line by line and be running in a minute.

Keep it where you look first#

Put the block where you already look when you open a project: the top of the README, or the project's state file next to the phase and next step. A document in a separate wiki is one you'll forget exists.

Let the agent keep it current#

Setup changes when the agent adds a dependency, a script or a service. Add one line to your end-of-session habit:

"If anything about how to run or test this project changed, update the How to run block."

And once in a while, try it on a fresh clone. If it doesn't work there, it doesn't work.

This is one more piece of the minimum process: small, written once, and saving time on every return.

Keep reading#

Frequently asked questions

Where should I document how to run a project?

In the repo, at the top of the README or in the project's state file — the exact commands in order, copy-pasteable, with any env file or port noted.

How do I stop forgetting the setup for old side projects?

Write the run commands down the first time you get the project running, and have your agent update them when the setup changes.

Can an AI agent figure out how to run a project?

Often, by reading package files and scripts — but it takes time and it can guess wrong. A written block is faster and removes the guess.

Sources

  1. Manage Claude's memory (CLAUDE.md) — Claude Code docs
  2. Claude Code overview — Anthropic docs

Found this useful? Share it on Reddit or pass it on.

RedditXLinkedInFacebook