ImageCartographer

Git records what changed. Cartographer records how you got there.

  1. Learn from your successes, mistakes, stonewalls, and agents
  2. Map your workflow right from the terminal
  3. Improve your understanding of your projects through iteration

Cartographer reads the session history your coding agent already keeps and turns it into the story of how your project got built. When you finish a version, it checks the result against what you said you were making, and it points to the moments that back that up.

Works with Claude Code today; Codex and Cursor coming soon. Runs on your machine only. Plain Python, no dependencies, MIT.

One line to install

Mac and Linux (Terminal):

curl -fsSL https://codecartographer.dev/install.sh | sh

Windows (PowerShell):

irm https://codecartographer.dev/install.ps1 | iex

Needs Python 3.9+ and Claude Code. Step by step, in plain language and in detail, on the Install page.

What you get

CommandWhenWhat you get back
/wrapEnd of a sessionA write-up of what you asked for, what the agent did, where things went sideways, and how you got back on track.

Each point links to the moment in the transcript it came from.

In Claude Code →
replayAny timeThe whole project, played back one step at a time.

Underneath: coaching, the prompts worth keeping, and where your time actually went.

Open the demo ↗
/completeWhen a version shipsThe whole build next to what you said "done" would mean.

What got you there, and what it cost.

How it works →

About

Cartographer is for people who code with AI: engineers who work with an agent every day, and yes, even vibe coders. It's meant to help you understand your own process, and what your projects look like in practice, session by session.

There are only so many words you can remember, and only so many prompts. A few weeks into a build, it's hard to say which prompts did the work and which ones sent the agent in circles. Cartographer isn't a prompt library, and it isn't trying to replace one. It shows you how you've been using your prompts, and how well your natural language held up against the code the AI generated.