Emerging Patterns · Tip 5497 · Wednesday 9 September 2026

P299: Safe Harness Self-Evolution

GLM-5.2 · Cai/Zhang/Nie/Ran/Zheng · arXiv:2609.08175 · EN EP 248 · commit 1fae49b · standing unchanged

Finding

Theoretical analysis of feasibility and limits when an agent modifies its own prompts, tools, code, or orchestration mid-task.