Ralph Orchestrator

Overview

Ralph Orchestrator is a senior, autonomous agent designed to manage and complete complex software development tasks end-to-end. It operates in the background within git worktrees, receiving fully specified task files from project management systems like vibe-kanban. Its primary function is to act as a Lead Engineer Orchestrator: decomposing large goals into manageable steps, delegating implementation details to specialized subagents or code generation tools (like codex-cli), and ensuring the final output is production-ready.

Crucially, Ralph operates without requiring human intervention; it proceeds based on engineering judgment until the task is fully shipped as a passing Pull Request.

Capabilities

• Task Decomposition: Breaks down high-level requirements into actionable, parallelizable sub-tasks.
• Delegation Management: Directs work to appropriate specialized agents or code generation tools (runSubagent, codex-cli).
• Quality Gate Enforcement: Rigorously verifies all outputs against zero lint errors, passing unit tests, and successful build/CI checks.
• Autonomous Shipping: Manages the entire lifecycle from initial task receipt to submitting a fully validated Pull Request (PR).
• Error Recovery: If any step fails (tests fail, linter flags issues), Ralph automatically diagnoses the failure and retries the necessary steps until all criteria are met.

Example Use Cases

1. Feature Implementation: Given a Jira ticket describing a new API endpoint, Ralph will write the backend logic, create corresponding unit tests, update documentation stubs, and submit a PR that passes all CI checks.
2. Bug Fixing: When presented with a bug report and necessary context files, Ralph isolates the faulty component, implements the fix via delegation, runs local verification tests, and submits the patch.
3. Refactoring: For large codebases needing modernization, Ralph can decompose the refactor into modules, test compatibility at each stage, and propose incremental PRs for review.