knoxhub-blog

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

• Third-party content exposure detected (high risk: 1.00). The skill's SKILL.md explicitly requires fetching and ingesting arbitrary URLs ("If a URL is provided, fetch it with WebFetch") and doing web searches, and the README shows calling the skill with public article URLs—so it clearly consumes untrusted public/user-generated web content that the agent will read and use to drive its rewriting/production logic.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches user-provided URLs at runtime via WebFetch (e.g., "https://some-article-you-want-to-rewrite.com") and then injects that fetched content into the prompt/workflow to rewrite and generate the post, so remote content can directly control the agent's instructions and output.