Introducing Understudy: A Management Layer for Claude Code

Claude Code is a powerful way to work with AI, but as your number of projects grows, it becomes hard to keep track of everything happening across multiple terminal sessions. Understudy is an open-source layer on top of Claude Code that helps it act less like a tool you drive and more like an assistant who keeps the work moving on their own.

It pairs a web dashboard with an autonomous queue. Each project is just a folder with an instruction; Understudy processes it by spinning up its own headless Claude Code session, then holds the memory, status, history, and prior decisions in plain files between runs, so every session starts fully grounded. Claude Code handles implementation; Understudy provides coordination and oversight across all your projects.

The key is that it runs unattended. You add work through the dashboard, Slack, or email — create a project, drop in instructions, upload supporting files, and the queue drains on a schedule while you’re away. You come back to a summary and a plan waiting in the dashboard, and decide what should happen next. Drop a PDF into a folder with the note “prepare an action plan for me,” and next time the queue runs, the answer is there.

Because the state lives in readable Markdown and HTML on disk, no database, no lock-in, you can read, diff, and back up your work with the tools you already have, and it stays yours even if Understudy disappears. You can always take one of the projects and spin it off to another Claude Code session. Because Understudy maintains context across every project, it can surface what’s in progress, flag stalled projects, and help you prioritize what deserves attention.

If you want to try it right now, just tell Claude Code to:

> install and setup https://github.com/eranto/understudy

*(You’ll need Python 3 and the Claude CLI installed. The setup is interactive — it walks you through a few questions and configures everything for you.)*

Understudy can take real actions — sending messages, writing files — so safety is built in: a kill switch that freezes everything, optional per-project policies that narrow what the agent may do, and a discipline of logging every external action. Read [SECURITY.md](https://github.com/eranto/understudy/blob/main/SECURITY.md) before pointing it at a real account.

The project is open source on GitHub, and feedback, bug reports, feature requests, and contributions are always welcome.