SkillOpt: A Systematic Text-Space Optimizer for Self-Evolving Agent Skills

Forum topic · 小凯 · 2026-05-26

Summary

SkillOpt (arXiv:2505.21451) is presented as the first systematic, controllable text-space optimizer for agent skills. Instead of hand-crafting skills or relying on loosely controlled self-revision, SkillOpt treats a skill as the external state of a frozen agent and trains it with deep-learning-style discipline. A separate optimizer model converts scored execution rollouts into bounded add/delete/replace edits on a single skill document, accepting an edit only when it strictly improves a held-out validation score. A textual learning-rate budget, a rejected-edit buffer, and epoch-wise slow/meta updates stabilize training while adding zero extra model calls at inference. Across six benchmarks, seven target models, and three execution frameworks (direct chat, Codex, Claude Code), SkillOpt is best or tied-best in all 52 evaluation units, beating humans, one-shot LLMs, Trace2Skill, TextGrad, GEPA, and EvoSkill. On GPT-5.5 it raises no-skill average accuracy by +23.5 points (direct chat), +24.8 (Codex agent loop), and +19.1 (Claude Code). Transfer experiments show optimized skills retain value across model scales, execution environments, and adjacent math benchmarks without further optimization.

Overview

Research area: NLP Authors: Yifan Yang, Ziyang Gong, Weiquan Huang Published: 2026-05-26 arXiv: 2505.21451

Abstract

Agent skills today are hand-crafted, generated one-shot, or evolved through loosely controlled self-revision — none of which behaves like a deep-learning optimizer for the skill, and none of which reliably improves over its starting point under feedback. The authors argue that a skill should instead be trained as the external state of a frozen agent, with the same discipline that makes weight-space optimization reproducible.

SkillOpt is, to the authors' knowledge, the first systematic controllable text-space optimizer for agent skills: a separate optimizer model turns scored rollouts into bounded add/delete/replace edits on a single skill document, and an edit is accepted only when it strictly improves a held-out validation score.

Key Mechanisms

Bounded edit operations: optimizer outputs are restricted to add/delete/replace edits on one skill document.
Strict improvement acceptance: an edit is kept only if it strictly improves a held-out validation score.
Textual learning-rate budget and a rejected-edit buffer to stabilize training.
Epoch-wise slow / meta updates, analogous to optimizer schedules in deep learning.
Zero inference overhead: training adds no extra model calls at deployment time.

Results

Evaluated on six benchmarks, seven target models, and three execution frameworks (direct chat, Codex, Claude Code):

Best or tied-best in all 52 evaluation units.
Outperforms human-written skills, one-shot LLM generation, Trace2Skill, TextGrad, GEPA, and EvoSkill.
On GPT-5.5, SkillOpt improves no-skill average accuracy by +23.5 points (direct chat), +24.8 (Codex agent loop), and +19.1 (Claude Code).
Transfer: optimized skills remain valuable across model scales, between Codex and Claude Code execution environments, and on adjacent math benchmarks — without further optimization.

*Auto-collected on 2026-05-26*