AutoDebug — 5-Agent Bug Detection Pipeline

AutoDebug is a multi-agent pipeline built entirely on Band that autonomously detects, analyses, and reports bugs in GitHub repositories — with a human always in the loop for the final fix. Five independent agents run as real Band-registered processes, coordinating through a shared Band chat room using @mentions. Each agent hands off accumulated context as JSON to the next, forming a clean pipeline: 1. DetectorAgent — clones the repo, finds failed CI runs via GitHub Actions API or runs Python files directly to surface errors 2. AnalyserAgent — uses AI to identify root cause, severity (low/medium/high/critical), and affected files 3. FixerAgent — generates a precise, minimal code fix with explanation using AI 4. TesterAgent — runs pytest to confirm the bug is real and reproducible 5. ResolverAgent — writes a clear, actionable fix report and delivers it by email The system never auto-merges code. Every fix is a report, not a commit — a human always reviews and applies the change. This human-in-the-loop design makes AutoDebug safe to run on production repositories. Built with Band SDK, NVIDIA API (MiniMax M2.7), GitHub API, Gmail SMTP, and pytest. Includes a CLI tool (autodebug start/stop/status/doctor) and a local fallback pipeline for offline testing.