Phoenix: Safe GitHub Issue Resolution via Multi-Agent LLMs

Abstract

Autonomous LLM agents can synthesize code changes at scale, but deploying them safely on real codebases requires guarantees that generated patches preserve correctness. Phoenix is a multi-agent LLM system that resolves GitHub issues from triage through pull-request creation, combining seven layered safety controls with a baseline-aware test evaluation strategy. Work is decomposed across six specialized agents — planner, reproducer, coder, tester, failure analyst, and PR agent — coordinated by a label-based GitHub webhook state machine, with every change checked against a baseline test run before a pull request is opened.