marketing-pricing

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

• Third-party content exposure detected (high risk: 1.00). The skill's Research Mode and Context C explicitly instruct the agent to fetch and scrape open web pages (e.g., agent-browser commands to open competitor pricing pages, G2/Capterra reviews, Reddit searches, and the Wayback Machine) and to audit user-provided live URLs, meaning untrusted third-party content is read and used to drive pricing decisions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's Research Mode instructs the agent at runtime to fetch live pages like "https://{competitor-domain}/pricing" via agent-browser (and even suggests installing agent-browser with "npm install -g agent-browser && npx playwright install chromium"), meaning external web content (e.g., https://{competitor-domain}/pricing) is fetched and directly used as source-of-truth to drive prompts and outputs and the install step pulls and executes remote code.