AgentDebugX is an open-source closed-loop debugger for LLM agents
UIUC-CS · hf · 2026-07-22
- AgentDebugX is an open-source toolkit for debugging LLM agents as a closed loop of Detect, Attribute, Recover, Rerun.
- Its key component, DeepDebug, performs multi-turn root-cause diagnosis using global trajectory understanding, structure-guided investigation, and cross-examination.
- On the Who and When benchmark, it reaches 28.8% exact agent-and-step accuracy on qwen3.5-9b, beating the strongest single-pass baseline at 21.7%.
- On GAIA, it repairs 13 of 73 failed tasks in one rerun, versus 4–6 for decoupled self-correction baselines, lifting overall accuracy from 55.8% to 63.6%.
- The project ships as a Python library, CLI, web console, and installable agentic skill, plus an opt-in Error Hub for sharing scrubbed failure bundles as debugging memory.
More from coding & agent
- OpenFPM CUDA-style kernels now run on Apple Silicon GPUs via Metal — Scobleizer · 2026-07-22
- Sol Pro built a 44-file handoff kit after a voice-first coding workflow — RileyRalmuto · 2026-07-22
- AI Dev Tools Zoomcamp opens with spec, context, loop, and graph engineering — Al_Grigor · 2026-07-22
- New paper defines four conditions that turn an LLM into a coding agent — alex_verem · 2026-07-22
- Obsidian’s 5 official agent skills teach Claude native notes, Bases and JSON Canvas — FinanceYF5 · 2026-07-22
- MCP lets Claude read and write Obsidian vault notes without copy-paste — FinanceYF5 · 2026-07-22