SkillCODER fingerprints LLM agent skills for copyright and buyer attribution
Original title: SkillCODER: Towards Auditing and Attribution of Copyright Infringement in LLM Agent Skills
Across three domains, six models and three frameworks, owner verification reached 99.2% and buyer Top-1 attribution 99.3%.
SkillCODER embeds buyer fingerprints in LLM agent skills so a defender can verify ownership from content alone. It parses each skill into SkillIR, places an Artifact-Gated Capsule on structural anchors, and encodes identity with controlled-vocabulary error-correcting codes. Paired probes recover owner and buyer identity while task fidelity is preserved. Signals held under paraphrase, compression, deletion, reorganization and five adaptive attacks.
- Conditional instructions are designed to activate only under verification queries.
- Ownership evidence stayed detectable under a method-aware sanitizer and multi-buyer collusion.
- Buyer attribution degraded gracefully under those adaptive attacks.
- The method outperformed prompt-level copyright auditing baselines.
From the original article
1 min read at arxiv.orgAgent skills package reusable workflows and task-specific procedures into deployable natural-language artifacts for LLM agents. Their scale and readability make them valuable but vulnerable to unauthorized redistribution and repackaging. Existing prompt copyright auditing methods struggle with long skills and introduce token-level perturbations that compromise readability.
By Enhao Huang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.