Autonomous engineering teams for code generation from GitHub issues. Zeroshot targets engineers and power users seeking to automate software development, delivering feature-complete, production-grade code by employing isolated, self-checking agents.

How It Works

The system uses a planner, worker, and validator agent architecture in isolated contexts, combating single-agent degradation and context burial. Validators perform "blind validation," assessing code without worker history. Rejections provide actionable feedback for iterative fixes, creating a robust accept/reject loop. This design prioritizes correctness and repeatability, with crash recovery via SQLite persistence.

Quick Start & Requirements

Install via npm: npm install -g @covibes/zeroshot. Prerequisites include Node 18+, Claude Code CLI (npm i -g @anthropic-ai/claude-code && claude auth login), and GitHub CLI (gh auth login). Supported on Linux/macOS; Windows WSL is not yet supported. Architecture details are in CLAUDE.md.

Highlighted Details

• Blind Validation: Objective code assessment without worker context or history.
• Repeatable Workflows: Agent count and model selection adapt to task complexity.
• Actionable Feedback: Rejections offer specific findings for targeted fixes.
• Crash Recovery: State persisted to SQLite for resumable workflows.
• Flexible Isolation: Options include no isolation, git worktree, or Docker containers.
• Cost Control: Configurable model ceilings prevent excessive API spending.

Maintenance & Community

Community support is available via Discord. Development guidelines, code of conduct, and security policies are detailed in CONTRIBUTING.md, CODE_OF_CONDUCT.md, and SECURITY.md.

Licensing & Compatibility

Licensed under the MIT license. No specific compatibility restrictions for commercial use are mentioned.

Limitations & Caveats

Windows WSL is not supported. Effectiveness relies on clearly defined acceptance criteria; unsuitable for exploratory tasks or vague objectives like general performance improvements without specific metrics.