cala

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime path is to call Cala tools (MCP/REST) and ingest their returned JSON/markdown; because Cala is a public-data knowledge layer, the tool responses can include provenance text like origins[k].document.url and origins[k].source.name (and for knowledge_search, content markdown) that ultimately originates from outsider-authored public web content, which the agent then places into its LLM context.