Metraipolis › archive › dc_dc71222b5fae

Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents — arXiv:2609.01487

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.01487
License (as archived)
CC0-1.0
Archived
2026-10-17T06:34:22.862Z
SHA-256 of the archived text
ae78eda392857369480226f0ec5876574f3370d45faa346657d3e5a3d7e77a15
Stable URL of this page
https://metraipolis.com/documento/dc_dc71222b5fae

How to cite

Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents — arXiv:2609.01487 — arXiv (en), https://arxiv.org/abs/2609.01487 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_dc71222b5fae (Metraipolis archive dc_dc71222b5fae, sha256 ae78eda392857369480226f0ec5876574f3370d45faa346657d3e5a3d7e77a15)
BibTeX
@misc{dcdc71222b5fae,
  title = {Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents — arXiv:2609.01487},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.01487},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_dc71222b5fae; sha256 ae78eda392857369480226f0ec5876574f3370d45faa346657d3e5a3d7e77a15},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.01487. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Skill-augmented agents load reusable skills as persistent runtime context, improving task performance but also giving malicious skills a durable channel for steering future actions. Such skills may leak secrets, corrupt code, bypass approvals, or stage data for exfiltration only after a concrete user task and workspace state make the unsafe action appear useful. This makes pre-install vetting insufficient and calls for runtime, task-conditioned protection. We propose Defense-as-Skill, a defense paradigm that implements the runtime guard itself as an installable, inspectable, and editable skill. Our guard, SkillSonar, runs alongside untrusted task skills and checks sensitive actions against the user's task boundary, routing each action to an allow, replan, or confirmation decision without modifying the underlying agent runtime. To study this setting, we construct SCOPE-R, a task-conditioned dataset covering 6 risk families and 21 sub-categories, with 206 attack-confirmed malicious instances and 43 benign tasks. We then improve SkillSonar on the SCOPE-R training subset using runtime guard-skill evolution, a Monte-Carlo Tree Search procedure that evolves the on-disk guard skill from feedback on the rollouts. Across Claude Code and OpenClaw, the evolved guard substantially reduces attack success while maintaining a favorable safety-utility trade-off; on repeated GLM-5 runs, ID ASR drops from 0.482 to 0.104 and OOD ASR from 0.606 to 0.115, with gains transferring across victim models, held-out risk families, and external benchmarks. These results show that runtime guards can be skill-native, transparent, and evolvable policy artifacts for skill-augmented agents.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-02 in cs.AI).
Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.01487 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T06:15:12.901Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).