Wreckit is a CLI tool that automates software development workflows by running an AI agent through your project backlog unsupervised.
The tool solves the problem of turning rough ideas into production-ready code by automating a multi-stage workflow. You provide a text file of half-baked ideas, and the agent progresses each item through five states: research, plan, implement, PR, and done. During research, the agent analyzes your codebase to understand patterns and conventions. During planning, it designs a solution broken into user stories with acceptance criteria. During implementation, it writes code story-by-story, running tests and committing as it progresses. Finally, it opens a pull request for your review. All work is stored as JSON and Markdown files in a `.wreckit/` directory, making the process inspectable, resumable, and git-trackable without requiring external databases or cloud services.
Wreckit suits teams comfortable with AI-assisted development who want to delegate backlog execution to an autonomous agent. It works best for projects where you can articulate ideas in text form and trust the agent to research your codebase and implement solutions. The tool requires LLM API access, which can be expensive given the agent's token consumption. You should expect to review and merge pull requests rather than run the tool in a fully hands-off manner.
The project shows active maintenance with regular updates to its core workflow and CLI commands. The tool includes debugging commands for individual phases, allowing developers to inspect and troubleshoot specific stages of the automation pipeline. Documentation covers both the essential commands for running the full workflow and phase-specific commands for investigating issues.
