Skip to content

SCAFFOLD: Toward Credential-Preserving Evidence Transfer for Reusable LLM Agent Skills

SCAFFOLD: Toward Warrant-Preserving Evidence Transport for Reusable LLM-Agent Skills

arXiv: 其他 AI 安全论文· arXiv:2611.23990· Nami Cho, Shailja Gupta·Indexed by this site · 13 hr ago· Originally publishedAlignment· Its place in the scan record →
AI summary

The paper proposes Warrantied Reuse Claims (WRC) as the unit of trust for LLM agent skill reuse, and gives WPET criteria for judging which historical evidence remains admissible after a context change. A WRC is a revocable authorization binding a versioned skill, a target context and evidence; WPET distinguishes evidence admissibility from claim truth, and specifies three-valued applicability and asymmetric transfer judgments. The paper contributes only a formal problem statement, design obligations and a reporting protocol, with no implemented verifier or empirical results.

Read originalarxiv.org