exa-search

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

• Third-party content exposure detected (medium risk: 0.30). references/web-search.md + scripts/exa_search.py ingest outsider-authored free text from the user’s runtime “query” (natural-language input) and pass it into Exa’s search_and_contents/search flow, which returns extracted contents that the skill then serializes.

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

• Potentially malicious external URL detected (high risk: 0.80). This skill calls Exa's content-fetching APIs at runtime to retrieve arbitrary user-supplied webpages and returns that page text verbatim (e.g., it fetches URLs like https://arxiv.org/abs/2401.04088), which effectively injects remote page content into the agent's output/context and can therefore control prompts.