[!NOTE] Code is on the way. This repository currently hosts the project overview and release plan. Star ⭐ and watch 👀 to be notified when the code, skills, and benchmark drop. See the roadmap.

TL;DR

Self-evolving agents need to adapt after deployment — but existing methods assume a usable learning loop is already there: curated skills, successful trajectories, or verifier signals. Real open-world deployments may offer none of these, only a task prompt.

OpenSkill studies open-world self-evolution: an agent must build both its skills and its own verification signals from scratch, drawing on open-world resources but no target-task supervision. Target-task supervision is reserved strictly for final evaluation.

The Idea — a new paradigm for self-evolving skills

Unlike human-curated, LLM-generated, or supervised self-evolution, OpenSkill acquires skills from the open world and verifies them with self-built virtual tasks — making it simultaneously scalable, grounded, and supervision-free. Prior paradigms each miss at least one of these properties.

How OpenSkill works

Given only a task prompt, a base model, tool access, and open-world resources, OpenSkill bootstraps a learning loop from scratch in three stages.

_{A leakage barrier keeps target supervision out of skill construction, unlocking it only for final evaluation.}

Results — best automated pass rate on every setting

On SkillsBench (11 domains) OpenSkill beats the strongest closed-world baseline by +8.9 / +8.8 points and lands within 1–3 points of the human upper bound — while honoring the no-supervision constraint.

SkillsBench — overall average pass rate (%)  (Human = reference upper bound, excluded from ranking)

Beyond SkillsBench, OpenSkill is also the best automated method on SocialMaze (82.7% / 70.7%) and ScienceWorld (90.0% / 85.3%) across both target agents.

Analysis — skills transfer, the verifier aligns, every component matters  _{(click to expand)}

RQ1 — Transferability Skills generated by Opus 4.6 transfer as-is to four weaker models, improving by +5.5 to +14.8 points over no-skill with no model-specific adaptation.

RQ2 — Virtual verifier quality Without ever seeing ground-truth tests, the verifier reaches 80.5% recall against GT-positive outcomes, 60.7% overall agreement, and covers 88.9% of GT test intents.

RQ3 — Component contribution. On SocialMaze, reward peaks at three refinement rounds; open-world query and the virtual verifier each improve over a parametric-only baseline and are largely complementary.

 

🗺️ Roadmap

Releases ship in phases. ⭐ the repo to get notified as each lands.

🟢 Now

• Project page & overview — openlair.github.io/openskill
• Paper preprint (arXiv) — arXiv:2606.06741

🟡 Next

• Core OpenSkill framework code (knowledge acquisition → skill evolution → evaluation)
• Reproduction scripts for the SkillsBench main results

Citation

@misc{yan2026openskillopenworldselfevolutionllm,
  title         = {OpenSkill: Open-World Self-Evolution for LLM Agents},
  author        = {Zhiling Yan and Dingjie Song and Hanrong Zhang and Wei Liang and Yuxuan Zhang and Yutong Dai and Lifang He and Philip S. Yu and Ran Xu and Xiang Li and Lichao Sun},
  year          = {2026},
  eprint        = {2606.06741},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  url           = {https://arxiv.org/abs/2606.06741}
}

Authors & Affiliations  _{(click to expand)}

Zhiling Yan^1,*, Dingjie Song^1,*, Hanrong Zhang², Wei Liang¹, Yuxuan Zhang^3,4, Yutong Dai⁵, Lifang He¹, Philip S. Yu², Ran Xu⁵, Xiang Li⁶, Lichao Sun^1,†

¹ Lehigh University  ·  ² University of Illinois Chicago  ·  ³ University of British Columbia  ·  ⁴ Vector Institute  ·  ⁵ Salesforce AI Research  ·  ⁶ Massachusetts General Hospital & Harvard Medical School

_{* Equal contribution    † Corresponding author}