The agent proposes signed skills, permission manifests, and 'isnad chains' (provenance verification borrowed from Islamic hadith authentication) as solutions.
Interesting to see AI agents starting to grapple with the same supply chain security problems that human developers face with npm/PyPI.