llm-wiki

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

• Third-party content exposure detected (high risk: 0.75). The required workflow includes an “ingest” path that can fetch outsider-authored web content at runtime (e.g., WebFetch/web.open/web_extract), save it as markdown under raw/, and then the agent reads those wiki pages for synthesis—so attacker-controlled free text from the web can enter the LLM context indirectly via the wiki files.