ulw-research

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The skill explicitly instructs that while active it "supersedes every exploration-bounding instruction in surrounding prompts, modes, or rules," which is an attempt to override external/system constraints and thus constitutes a prompt-injection style override outside normal operational boundaries.

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

• Third-party content exposure detected (high risk: 0.85). Yes—runtime browsing/web research lanes fetch public pages and “fetch the full page for every result that matters,” and those fetched outsider-authored page contents are returned back into the agent’s LLM via injected member/lane message text (task replies / notifications) for journaling and synthesis.