zeroshot

zeroshot is an open-source meta-harness for coding agents that runs a configurable graph of workers and reviewers instead of a single agent checking its own work. It is built for senior engineers, teams automating PR workflows, platform teams standardizing agentic processes, open-source maintainers, and AI power users who want verification rather than a single agent's self-assessment.

A worker agent produces a change, then independent reviewer agents, started in fresh sessions and optionally on a different model, check the work against a graph defined for the repository; a failing check sends the reason back to the writer, and the fix goes through every review again until it passes. The graph also decides where finished work goes: kept local, pushed as a branch, opened as a PR, or merged through CI.

zeroshot ships skills for Claude Code, Codex, Copilot CLI and Pi that design a graph from a repo's build, test and CI setup, turn an idea or issue into a plan with checkable acceptance criteria, and start, follow or troubleshoot a run. Every event lands in a durable ledger so a crashed run can resume, and a browser UI (zeroshot ui) lets a user watch a run live.

The zeroshot CLI is released under the MIT license and installs with npm install -g @the-open-engine-company/zeroshot, installing as a skill into Claude Code, Codex, Copilot CLI and Pi. Runs execute locally using a user's own model login and subscriptions, or on a self-hosted Docker target, or through zeroshot Platform, which runs work in parallel off a user's own machines using their own model API keys.

Use cases

• Gating a pull request with independent security and code review checks
• Standardizing an agentic workflow across an infrastructure or platform team
• Working through an open-source project's issue backlog with verified changes