Research map / Large language models / Agents, code and math
AgentXRay: White-Boxing Agentic Systems via Workflow Reconstruction
Ruijie Shi, Houbin Zhang, Yuecheng Han, Yuheng Wang, Jingru Fan, Runde Yang, Yufan Dang, Huatao Li, Dewen Liu, Yuan Cheng, Chen Qian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| remove tools from primitive space (ablation) | AgentXRay · average of 5 domains | SFE similarity: 0.356 -> 0.426 | Improved |
| remove tools from primitive space (ablation) | AgentXRay · MetaGPT | SFE similarity: -46.0% | Improved |
| add Red-Black Pruning to MCTS (ablation) | AgentXRay (Selected Tools) · ChatDev | SFE similarity: 0.286 -> 0.509 | Improved |
| add Red-Black Pruning to MCTS (ablation) | AgentXRay · average of 5 domains | SFE similarity: 0.318 -> 0.421 | Improved |
| add Red-Black Pruning to MCTS (ablation) | AgentXRay (Selected Tools) · ChatDev | token consumption: 17.16M -> 13.33M | Improved |
| add Red-Black Pruning to MCTS (ablation) | AgentXRay · search depth | final workflow length: L=2 -> L=6 | Improved |
| add Red-Black Pruning to MCTS (ablation) | AgentXRay · MetaGPT | SFE similarity: 0.334 (w/o) vs 0.470 (Selected) | Mixed |
| add Red-Black Pruning with all tools on Atoms (ablation) | AgentXRay · Atoms open-weight | similarity: 0.328 -> 0.326 (Selected), 0.337 (All Tools) | Mixed |
| use workflow search over primitives instead of SFT | AgentXRay · average of 5 domains | SFE similarity: 0.196 -> 0.426 | Improved |
| use stronger single model Claude Opus 4.5 with ReAct instead of structured search | Claude Opus 4.5 · average of 5 domains | SFE similarity: 0.299 vs 0.426 | Hurt |
| use AgentXRay instead of AFlow | MCTS workflow search · average of 5 domains | SFE similarity: 0.339 -> 0.426 | Improved |
| use AgentXRay instead of AFlow with open-weight primitives | MCTS workflow search · Atoms | similarity: 0.321 -> 0.337 | Improved |
Closest papers by meaning
- FlowSteer: Towards Agents Designing Agentic Workflows via Reinforced Progressive Canvas Editing (NeurIPS 2026 · similarity 0.89)
- AFlow: Automating Agentic Workflow Generation (ICLR 2025 · similarity 0.89)
- Benchmarking Agentic Workflow Generation (ICLR 2025 · similarity 0.89)
- AgentSwift: Efficient LLM Agent Design via Value-Guided Hierarchical Search (AAAI 2026 · similarity 0.89)
- GraphFlow: A Graph-Based Workflow Management for Efficient LLM-Agent Serving (ICML 2026 · similarity 0.89)
- Flow: Modularized Agentic Workflow Automation (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
