---
title: "Bootstrapping Composer with autoinstall"
description: "RL training depends on runnable environments, because broken setup can consume tokens and compute without producing a reward signal or can make a task unsolvable. Composer autoinstall addresses this by using earlier Composer models to turn unconfigured repository checkouts into working mock environments. A Cursor agent first proposes 10 commands and expected outputs, while a separate Composer agent configures the checkout, mocks dependencies, and is evaluated against three selected commands. The process can create placeholder files, images, database tables, MinIO configurations, Docker containers, and startup scripts, restarting up to five times before discarding an environment. In a Celo monorepo experiment, the agent installed dependencies including Foundry and eventually mocked a user to run a minimal application; Composer 2 later scored 61.7% on Terminal-Bench versus 47.9% for Composer 1.5."
---

# Bootstrapping Composer with autoinstall

[Cursor](https://yomu.fyi/company/cursor) · Shomil Jain, Joshua Warner & Andrew Zhai · May 6, 2026

**Type:** Explainer

## Summary

RL training depends on runnable environments, because broken setup can consume tokens and compute without producing a reward signal or can make a task unsolvable. Composer autoinstall addresses this by using earlier Composer models to turn unconfigured repository checkouts into working mock environments. A Cursor agent first proposes 10 commands and expected outputs, while a separate Composer agent configures the checkout, mocks dependencies, and is evaluated against three selected commands. The process can create placeholder files, images, database tables, MinIO configurations, Docker containers, and startup scripts, restarting up to five times before discarding an environment. In a Celo monorepo experiment, the agent installed dependencies including Foundry and eventually mocked a user to run a minimal application; Composer 2 later scored 61.7% on Terminal-Bench versus 47.9% for Composer 1.5.

## Context

RL training requires runnable environments. If setup is broken, a model may spend tokens debugging instead of solving problems, and an unsolvable environment can waste compute without producing a reward signal.

## Approach / What changed

Autoinstall uses two agents and two stages: a Cursor agent proposes commands and expected outputs, then a separate Composer agent configures the repository, mocks missing dependencies or services, and is tested against three selected commands. The process is retried up to five times before an unsuccessful environment is discarded.

## Takeaways

- The first stage proposes 10 setup, test, or launch commands and describes the output expected from a correctly configured environment.
- Autoinstall can create mock files, images, database tables, MinIO configurations, Docker containers, and startup scripts for missing project dependencies or long-running processes.
- In the described benchmark, Composer 2 scored 61.7% on Terminal-Bench, compared with 47.9% for Composer 1.5.

**Tags:** [AI Agents](https://yomu.fyi/topic/ai-agents), [Reinforcement Learning](https://yomu.fyi/topic/reinforcement-learning), [Testing](https://yomu.fyi/topic/testing)

- Source: [Cursor](https://cursor.com/blog/bootstrapping-composer-with-autoinstall)
- Source URL: https://cursor.com/blog/bootstrapping-composer-with-autoinstall
- Ingested by Yomu: 2026-09-02T11:43:16.669Z

[Read original post](https://cursor.com/blog/bootstrapping-composer-with-autoinstall)