Ralph Wiggum. Cloud Scale.
Autonomous AI agent orchestration for your codebase. Run headless in GitHub Actions or interactively via TUI — powered by the Copilot SDK.
Run without TUI for CI/CD automation. Includes a Docker-based GitHub Action for scheduled agent swarms.
Multi-model support (Copilot, Llama, GitHub Models) with optional gastown cloud runtime offloading.
Drop-in GitHub Action for headless runs. Schedule agent swarms on cron, dispatch manually, or trigger from events.
Generate structured task plans from natural-language goals. Just describe what you want and go.
Built-in critic evaluates agent output against the task before tests run, refining quality automatically.
Consecutive failures trigger automatic task restructuring — split, rewrite, or inject prerequisites.
Pass structured outputs from one task as inputs to the next. Design docs flow into code tasks automatically.
Full audit trail of every agent execution — model, prompts, diffs — exportable as openclaw JSON.
Beautiful terminal UI showing tasks, progress, and real-time logs with pause/resume controls.
Max iterations, intelligent multi-factor scheduling, parallel phase execution, and git-backed reversibility.
# Add wreck-it to your GitHub Actions workflow - uses: randymarsh77/wreck-it/action@main env: COPILOT_API_TOKEN: ${{ secrets.COPILOT_API_TOKEN }} # Or run locally with the TUI git clone https://github.com/randymarsh77/wreck-it.git cd wreck-it && cargo build --release wreck-it plan --goal "Build a REST API with auth" --output tasks.json wreck-it run --task-file tasks.json --max-iterations 50 # Run headless for CI/CD pipelines wreck-it run --headless --task-file tasks.json --max-iterations 100