Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution

arxiv:2608.08311

Original source · Paper JSON

Agents, open-endedness, AGI · 2026-08-08

Cites (4) · Cited by (0)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2

Contents

• Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution
• Abstract
• 1 Introduction
• Contributions.
• 2 Related Work
• Self-evolving agents.
• Harnesses and coding agents.
• Persistent memory and deployment.
• Benchmarks and protocol validity.
• 3 Ouroboros Architecture
• Commit pipeline.
• Task outcomes and verification.
• Operational identity and memory.
• Two modes of core evolution.
• Benchmark execution and evidence.
• Subagents and patch integration.
• 4 Hope: Free Evolution under Human Interaction
• Multi-channel state.
• Social-interaction-driven development.
• Controls on self-directed work.
• Operator boundary.
• Evolution during deployment.
• 5 Evaluation
• Terminal-Bench 2.1.
• OSWorld-Verified.
• CL-Bench.
• SWE-bench Pro and GAIA.
• 6 Trajectory Audits and Harness Improvements
• Reward hacking.
• Contamination.
• Isolation failure.
• Remote-state drift.
• Continual-memory failures.
• 7 Operational Safety Controls
• Risk: agents that choose their own model APIs.
• Guardrails in use.
• Observed behavior.
• 8 Conclusion
• Limitations
• Ethical Considerations
• Use of AI Assistance
• Acknowledgments
• Appendix A Guardrails in Full
• Appendix B Constitution (Abridged)
• Appendix C Benchmark Configuration Disclosure
• Appendix D Supplementary Figures and Tables
• Instructions for reporting errors