Red Queen

Red Queen is a deterministic orchestrator for AI coding workflows. It shuttles tickets through a fixed sequence of phases — spec, plan review, code, review, test, and human review — and is described as a pipeline for people who already have a coding agent. The system is built to route work without spending AI tokens on coordination, so the routing layer is pure deterministic logic rather than an LLM-driven decision process.

Its workflow is configured in YAML, where phases, skills, and gates can be defined and changed. Each phase runs in isolation with a purpose-built prompt, including roles such as spec writer, coder, reviewer, tester, and comment-handler. Human-in-the-loop review checkpoints are first-class parts of the configuration, and they can be added, removed, or reordered. Failed phases retry up to three times before escalation to a human gate, and the workflow can be extended with custom review checkpoints such as a security review gate.

Red Queen is issue-tracker native. It supports bidirectional sync with Jira and GitHub Issues, and work moves from the tracker through the pipeline and back again. The orchestration layer also supports an optional webhook server for instant response, along with polling reconciliation that works out of the box, and it does not require an external queue. The package is installed with npm using npm install -g redqueen, and the quickstart describes initialization inside a repository, local startup, and a live dashboard at 127.0.0.1:4400. It also dispatches Claude Code as CLI subprocesses, and the page notes that the CLI must be logged in.

The project presents itself as complementary to coding agents rather than a replacement for them. It says Red Queen sits above the agent, does not host or manage the agent’s skills, and is not an agent runtime, a generic skill registry, a spec framework, or an interactive coding tool. The page also describes adapter-based support for integrations, with each integration implementing an IssueTracker or SourceControl interface.