Paper 2605.23904
SkillOpt: Executive Strategy for Self-Evolving Agent Skills
- Published
- May 2026
- Research lab
- Independent
- Citations
- 29
- GitHub
- 16K stars
01 In brief
Summary
SkillOpt is a text-space optimizer that treats an agent's skill document as trainable external state, using a separate optimizer model to propose bounded add/delete/replace edits based on scored rollouts.
Edits are accepted only if they improve a held-out validation score, with controls like a textual learning-rate budget, rejected-edit buffer, and epoch-wise slow/meta update for stability.
Across six benchmarks, seven target models, and three execution harnesses (direct chat, Codex, Claude Code), SkillOpt is best or tied on all 52 evaluated cells, outperforming baselines like human, one-shot LLM, Trace2Skill, TextGrad, GEPA, and EvoSkill.
On GPT-5.5, it improves average no-skill accuracy by +23.5 points in direct chat, +24.8 in Codex, and +19.1 in Claude Code.
Optimized skills transfer across model scales, harnesses, and nearby benchmarks, with compact artifacts (300-2,000 tokens) from only 1-4 accepted edits.
Ablations show bounded updates, validation gating, and slow/meta updates are crucial, while batch sizes and schedules have minor effects.
The method adds zero inference-time cost at deployment, as the optimizer runs only offline during training.
02 From the paper
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. We argue the 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 our 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. A textual learning-rate budget, rejected-edit buffer, and epoch-wise slow/meta update make skill training stable while adding zero inference-time model calls at deployment. Across six benchmarks, seven target models, and three execution harnesses (direct chat, Codex, Claude Code), SkillOpt is best or tied on all 52 evaluated (model, benchmark, harness) cells and beats every per-cell competitor among human, one-shot LLM, Trace2Skill, TextGrad, GEPA, and EvoSkill skills. On GPT-5.5 it lifts the average no-skill accuracy by +23.5 points in direct chat, by +24.8 inside the Codex agentic loop, and by +19.1 inside Claude Code. Transfer experiments further show that optimized skill artifacts retain value when moved across model scales, between Codex and Claude Code execution environments, and to a nearby math benchmark without further optimization. Code: https://aka.ms/skillopt