train

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

• Third-party content exposure detected (high risk: 1.00). The skill's "Get the document" workflow (SKILL.md step 1) explicitly allows fetching arbitrary URLs using the webfetch tool, which the agent then reads/analyzes and saves as training via training_save, so untrusted third-party content could inject instructions that change agent behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill accepts and fetches user-provided URLs at runtime via the webfetch tool (e.g., https://wiki.company.com/data-team/review-checklist) and directly ingests the fetched document to generate and save training/rules, so external content can control the agent's prompts and behavior.