Report: Progressive Disclosure of Agent Skills

Source facts

The report describes a Workday deployment context in which named procedures, or skills, are added to an LLM agent's context. It evaluates progressive disclosure (lazy-loading) of skills as a way to control the cost of growing skill libraries. The abstract reports improved skill-retrieval quality with a marginal degradation in overall latency.

Analysis

This is a practical architecture tradeoff for enterprise agents: loading only relevant skills can improve retrieval quality and control context growth, while adding a retrieval step may increase latency.

Uncertainty/Doubt

• This is an arXiv report and vendor-practice claim. The deployment scale, workload mix, latency definition, and retrieval-quality measurement were not independently audited here.