first-principles-coach

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

• Third-party content exposure detected (medium risk: 0.30). Phase 1 instructs that if the user provides a URL, the workflow uses WebFetch to read that page text “to 파악한 뒤” inform subsequent questioning, creating a runtime path where outsider-authored free text from an arbitrary user-submitted URL can be ingested.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill uses WebFetch to read a user-provided website ("웹사이트 주소") at runtime and injects that page's content into Phase 1 questioning, so external content can directly influence agent prompts.