summarize

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

• Third-party content exposure detected (high risk: 0.90). The to-markdown.mjs script (and SKILL.md) explicitly accepts arbitrary URLs and calls runMarkitdown which runs uvx --from 'markitdown[pdf]' markitdown to fetch/convert public web pages/PDFs, then passes the full converted Markdown into summarizeWithPi (embedded in the LLM prompt), so untrusted third‑party content from the open web can be interpreted and influence the model's outputs.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill runs uvx --from 'markitdown[pdf]' markitdown <url> on any user-provided https:// URL at runtime and then embeds the fetched Markdown into the prompt passed to the pi command, so arbitrary external HTTPS content (e.g. any https://... user-supplied URL) directly controls the agent's prompt.