web-research

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.80). This is a personal GitHub Releases page (unknown/individual account) that likely hosts binary releases — downloading and running executables from such repositories is risky unless the author, release artifacts, checksums/signatures, and community trust are verified.

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

• Third-party content exposure detected (high risk: 0.95). The skill fetches public web pages at runtime (Step 1: browser39 search + browser39 batch to retrieve arbitrary URLs, then writes extracted markdown into docs/web/... which is later read by tools like get_symbols/get_context), so outsider-authored free text from the web can enter the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly runs browser39 at runtime to fetch arbitrary remote pages via the JSON command's "url":"THE_URL" (e.g., any https://... URL), and the fetched markdown is indexed and used as agent context so external pages can directly influence/ control prompts.