openclaw-serper

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly performs Google searches and then fetches and extracts full page text from each result URL (described in SKILL.md and implemented in scripts/search.py via trafilatura), exposing the agent to arbitrary public web pages whose untrusted content the agent is expected to read and act on.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill makes runtime POST requests to https://google.serper.dev/search and https://google.serper.dev/news and then fetches arbitrary result pages via trafilatura.fetch_url, injecting the extracted remote page content into the agent output (model context), so remote content can directly control prompts.