Verdict
Built from two strong, well-sourced reports. The agent-ready thread's technical claims are grounded in primary standards and major papers (arXiv, RFCs, spec sites); the federated thread in framework docs, FL literature and UK guidance.
Open items carried from the sources
- Context Hub (now resolved): the primary source is Andrew Ng's open registry Context Hub (
andrewyng/context-hub, MIT). It confirms the architectural point — a content/skills registry (curated, versioned docs and agent skills, with annotations-as-untrusted and author feedback) is distinct from MCP transport and A2A interoperability. - Federated LLM fine-tuning evidence in regulated UK settings is thinner than classical FL; named UK local-authority FL deployments are sparse.
Conventions
Glossary terms are linked from the concept that uses them; relationship direction is meaningful; see Linking conventions.
Terms
Glossary terms used here: Agent skill.