Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams

arxiv:2606.01770

Original source · Paper JSON

Harness · 2026-06-01

Cites (2) · Cited by (0)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2
3. Part 3
4. Part 4

Contents

• Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams
• Abstract
• 1 Introduction
• 2 Related Work
• 3 Method
• 3.1 Problem Formulation
• 3.2 Analytical Framework
• Proposition 1 (Regret Decomposition) .
• 3.3 Sustained Auto-Harness via Multi-Agent Evolution
• 3.4 Solve-Time Adaptation via Harness-Tree Routing
• 3.5 Human-in-the-Loop Channel
• 4 Experiments
• 4.1 Experimental Setup
• 4.2 Comparison with Baselines
• 4.3 Benchmark Bottlenecks
• 4.4 Stateful Multi-Agent Evolution
• 4.5 Solve-Time Routing on the Harness Tree
• 4.6 Human Steering for Auto-Harnessing
• 5 Conclusions
• 6 Limitations
• 7 Ethics Statement
• 8 AI Usage Statement
• Appendix A Benchmark and Evaluation Details
• A.1 PolyBench
• A.2 CTF-Dojo
• A.3 FutureX
• Appendix B Implementation and Reproducibility Details
• Appendix C Benchmark Non-Stationarity Diagnostics
• Appendix D Further Experiments
• Appendix E Per-Domain and Per-Category Breakdowns
• Appendix F Multi-Agent Evolution Dynamics
• Appendix G Routing Behaviour and Branch Performance
• Appendix H Human-in-the-Loop Event Log
• Appendix I Run-Detail Analysis
• Appendix J System Prompts
• Instructions for reporting errors