search-agentized-websites

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The runtime path GET <origin>/agents → POST <origin>/agents/search → GET <origin>/agents/resources/{id} ingests free text fields like query, matches.excerpt, and retrieved text/markdown from a publisher-controlled Agentize interface that an outsider can point to via an origin they choose.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill's client fetches arbitrary site origins at runtime (e.g., https://example.com) to retrieve manifest, search results, and resource Markdown which are injected into agent responses, so remote content can directly control prompts/answers.