SkillOpt: Turning Agent Skills into Trainable External Memory

A new paradigm treats agent skills not as brittle prompts or expensive fine-tuning targets, but as an optimizable external text state — exactly like model parameters in deep learning.

SkillOpt introduces a systematic optimizer for natural-language skills: bounded edits, validation gates, meta-updates, and a clean separation between skill storage and execution. The result is compact, transferable, interpretable procedural rules that improve agent performance with zero additional inference cost.

Tested across six benchmarks, seven models, and three execution harnesses, SkillOpt consistently outperforms human-written skills and all baselines.

This Ledger Entry expands how readers think about agent improvement by moving from fragile prompt engineering and costly retraining to a stable, scalable, external skill-optimization layer — a foundational shift for self-evolving agents.