deep-research

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

• Third-party content exposure detected (high risk: 0.85). The workflow’s runtime LLM context is populated by subagents’ web_search/web_fetch results (public web pages/articles) that are written into task-*.md notes and then read by the lead agent in P3/P4/P5, which is outsider-authored free text from public web content.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)