zeroshot

zeroshot is an MIT-licensed meta-harness for coding agents. Its argument is that a skill can ask for a review but the agent decides whether to listen, so zeroshot puts coding, independent review and repair into a workflow graph whose review gates and retry limits are enforced by the engine. You choose the agents and models, define what must pass, and decide where rejected work goes; each agent can run on its own model and harness, including Pi, Claude Code, Codex and Copilot.

The project reports that on 20 ProgramBench tasks a small model in a zeroshot loop matched a larger model's score at a median 58% of its cost; that is the project's own result. Install with npm install -g @the-open-engine-company/zeroshot, which fetches a verified native binary for Linux, macOS or Windows (Node.js 18+) and adds agent skills for Codex, Copilot, Pi and Claude Code. Local runs reuse your existing harness login, including subscriptions. Runs can be local, self-hosted with Docker, or on the hosted Zeroshot Platform.