agent-deepweb

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

• Third-party content exposure detected (high risk: 0.75). SKILL.md describes runtime fetching of arbitrary authenticated URLs via agent-deepweb fetch/graphql, whose stdout includes a JSON envelope with body (decoded response) and headers (redacted) that the agent will read into its LLM context; if the remote server is an outsider, its response body is outsider-authored free text.