Agent ecosystems are becoming a runtime supply-chain attack surface

The placeholder third-party[.]com campaign reportedly delivered tailored malicious content across more than 1,700 repositories, exposing static-analysis blind spots in skills, documentation, links, and MCP-like workflows. New reporting on malicious reverse-shell skills, model-loading risks, and low-volume poisoning reinforces the need for provenance, runtime link tests, content-variation testing, sandboxing, and instruction-integrity checks.