Progressive Skill Discovery as Access Control for Tool-Using LLM Agents: Structural Governance through Role-Scoped Capability Delivery
3.80T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2609.28693.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryThe paper introduces skilder, a framework that packages LLM agent capabilities into roles (skills, tools, instructions, and limits) delivered via an MCP server. Agents progressively discover required roles and only receive tools scoped to learned skills, enabling deterministic authorization. Evaluation across 13 tasks and 6 models shows unauthorized calls are blocked while flexibility is preserved.
Why it mattersTargets a real enterprise agent gap: prompt policies lack hard enforcement. Role-scoped capability delivery is a testable structural approach, not a handwave.
Cited by
No citations on record.