SkillOpt: Executive Strategy for Self-Evolving Agent Skills

Authors: Yifan Yang, Ziyang Gong, et al.

Breakthrough: Introduces the first systematic controllable text-space optimizer for agent skills. Treats skills as external state of a frozen agent, using a separate optimizer model to perform bounded edits on skill documents based on scored rollouts.

RSI Signal: High. This operationalizes the yanhua.ai mission of treating the agentic layer (harness + skills) as the primary unit of evolution rather than weight-space fine-tuning.

From Raw Experience to Skill Consumption: A Systematic Study of Model-Generated Agent Skills

Authors: Zisu Huang, Jingwen Xu, et al.

Breakthrough: Comprehensive study of the full skill lifecycle. Identifies that strong extractors are often weak consumers, causing "negative transfer." Proposes a meta-skill to guide extraction toward utility.

RSI Signal: Medium-High. Essential for stabilizing recursive loops where agent performance can degrade if "bad skills" are incorporated into the library.

CHRONOS: Temporally-Aware Multi-Agent Coordination for Evolving Data Marketplaces

Authors: Joydeep Chandra

Breakthrough: Three-layer architecture for dynamic data environments using neural-ODE temporal decay for knowledge graphs and changepoint-conditioned Shapley pricing.

RSI Signal: Medium. Relevant for agents managing long-term memory evolution and decentralized value attribution in multi-agent swarms.