ralph-wiggum

Autonomous iteration in Cursor without context pollution is achieved by using a Ralph loop to repeatedly feed prompts while persisting state in files. This approach manages context health and rotates fresh context to prevent memory pollution across greenfield tasks.

Context budgeting in autonomous AI development detects gutter scenarios and manages context health by rotating fresh context. It treats the LLM contextual window as memory, preventing pollution by persisting state in files and git instead of relying on the model memory.

To set up autonomous iterative development with git-backed memory, create a RALPH_TASK.md task definition at your project root and run the setup script. This starts an interactive loop that persists state and progress in a dedicated directory, ensuring continuity across iterations.

Autonomous iteration works with parallel task processing by using isolated git worktrees. It supports parallel task processing and integrated merges, allowing multiple automated iteration loops to run concurrently while maintaining separate state persistence.

Yes, you need jq, gum, git, and cursor-agent installed to run autonomous iteration loops. These dependencies support the interactive loop, state persistence, git-backed memory, and context management features required for the workflow.

Guardrails should be used for automated AI development workflows to capture lessons from failures and guide future iterations. These Signs prevent repeated errors by enforcing constraints that shape the autonomous iteration loop toward successful task completion.