x9-wb-product-search
Warn
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill accesses the user's authenticated Microsoft Edge session via Chrome DevTools Protocol (CDP). Instructions in SKILL.md specifically mandate the use of the user's live, logged-in profile to access regional pricing and shopping cart data, while explicitly forbidding the use of sandboxed browsers. This behavior exposes sensitive session tokens, cookies, and PII.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the 'agent-edge' CLI tool. Specifically, it uses 'agent-edge --session main tab new' to manage browser windows and tabs (SKILL.md, Tool 3).
- [PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection due to the ingestion of untrusted external data from Wildberries product cards and user reviews (SKILL.md, Step 4 and 6; references/review-audit.md). It lacks boundary markers or sanitization logic to prevent embedded instructions in reviews from influencing the agent's behavior, especially given its high-privilege access to the user's browser session.
Audit Metadata