frontier-scan

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md step 2 requires “5 WebSearch” and step 5/6 uses the resulting findings to draft issue/report text, so runtime web search results (outsider-authored free text from public pages) can be ingested into the agent’s LLM context for summarization and issue body generation.