vendor-termination-data

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

• Third-party content exposure detected (medium risk: 0.65). The required runtime workflow ingests vendor-provided free-text artifacts (e.g., deletion certification fields like “review_notes”/“exceptions” and validation “notes”) into the agent/LLM context via the vendor’s written deletion certificate and return/validation communications, which are outsider-authored.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The prompt specifies destructive, system-level data deletion actions (e.g., DROP/DELETE + VACUUM, secure overwrite, crypto-erase, bucket deletion, degaussing) that modify or destroy machine state and typically require elevated/sudo privileges and access to system resources, so it should be flagged.