# zeroshot

> Open-source CLI that runs coding agents through an enforced graph of worker, independent reviewers, and repair loops before delivery.

zeroshot is an open-source npm CLI from The Open Engine Company that adds an enforced QA process around coding agents such as Claude Code, Codex, and GitHub Copilot. A worker agent implements a task, independent reviewer agents check the result, and rejected work goes back through a bounded repair loop before anything is delivered. The latest release listed on GitHub is v10.10.1.

## What It Is

zeroshot does not replace Claude Code, Codex, or Copilot; it runs one of them as the worker and as the separate reviewers. You define the process as a graph that sets the sequence, parallel steps, retry paths, and exit conditions before execution starts. Runs are bounded, and events are recorded in a durable SQLite ledger. The project states that a passing run means its configured checks accepted the work, and that coverage depends on the requirements, reviewers, tests, and environment you supply.

## How the built-in graph works

The built-in `software-change` template follows this flow:

1. A worker implements the task in the current Git worktree.
2. Acceptance and code reviewers check the result independently, in parallel.
3. Rejected work goes to one repair worker with both reviews' feedback, then through both reviews again.
4. With delivery enabled, accepted work proceeds through the configured Git and CI steps, and delivery conflicts return through repair and review.

Delivery is opt-in: with no flag the work stays local, `--push` pushes the managed branch, `--pr` prepares a mergeable pull request, and `--ship` proceeds through PR, CI, and merge.

## Custom graphs and runtimes

Users can add their own stages, such as a bugfinder that writes adversarial tests or end-to-end tests, and choose each node's model, effort, and instructions. A per-node runtime config lets reviewers use a different model from the worker. A graph plus runtime settings can be saved as a profile shared between the CLI and the browser UI. A separate built-in `auto-research` template runs ten bounded experiment iterations by default, with scouts, a planner, independent judges, and an auditor, keeping results under `.zeroshot/research`.

## Setup path

Install with `npm install -g @the-open-engine-company/zeroshot`, which needs Node.js 18 or newer and installs a native binary for Linux x64/arm64, macOS x64/arm64, or Windows x64. Local runs reuse an existing Codex, Claude Code, or GitHub Copilot login, including subscription-backed sessions, and other providers can be configured with API keys. `zeroshot ui` opens a local workspace UI for viewing runs, and graphs can also run on a self-hosted Docker target. A Python SDK and a cluster API are also available, and the site describes a separate hosted zeroshot Cloud option.

## Features
- Enforced worker, independent review, and repair loop
- Built-in software-change graph with parallel acceptance and code review
- Custom graph topologies with user-defined stages
- Per-node model and effort configuration
- Reusable profiles shared between CLI and browser UI
- Opt-in delivery via --push, --pr, and --ship
- Auto-research template with bounded experiment iterations
- Local runs reusing existing harness logins
- Self-hosted Docker target
- Durable SQLite event ledger
- Restart or resume failed runs
- Python SDK and cluster API

## Integrations
Claude Code, Codex, GitHub Copilot, OpenAI, Anthropic, OpenRouter, Amazon Bedrock, GitHub, Docker

## Platforms
WINDOWS, MACOS, LINUX, WEB, API, JETBRAINS_PLUGIN, DEVELOPER_SDK, CLI

## Pricing
Open Source, Free tier available

## Version
v10.10.1

## Links
- Website: https://zeroshot.sh
- Documentation: https://zeroshot.sh/docs/current/
- Repository: https://github.com/the-open-engine/zeroshot
- EveryDev.ai: https://www.everydev.ai/tools/zeroshot