exa-search

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested when the skill runs Exa web search and/or URL extraction with --text/--highlights, because Exa returns scraped webpage content (e.g., results[].text / results[].highlights) from public third-party sources into the JSON that the agent then uses as LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's extractor fetches arbitrary external URLs at runtime via Exa's /contents API (e.g., URLs like https://arxiv.org/abs/2401.04088 passed to scripts/exa_extract.py) and returns their content verbatim for inclusion in responses, which can directly inject remote text into the agent's context and thus influence prompts/behavior.