thaki-AI/daily-paper-2026-09-21-registry-hijack-skill-router-robustness

The Registry Hijack: Measuring Adversarial Description Robustness of Hybrid Skill Routers in a 2,000-Skill Agent Ecosystem TL;DR — The nightly LLM curator that rewrites the SKILL.md descriptions of a 2,029-skill registry makes those descriptions - the sole routing signal of a hybrid BM25+embedding router - a routine write target, so one crafted or drifted description can hijack top-1 routing from gold skills while Recall@5 moves by only ~0.2 points (vs ~49 for top-1); we show… See the full description on the dataset page: https://huggingface.co/datasets/thaki-AI/daily-paper-2026-09-21-registry

<iframe src="https://silytics.si/embed/item/7b430976-227b-4307-9ca4-49f7e5aa6118" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.