temu-review-scraper
Fail
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONPERSISTENCEEXTERNAL_DOWNLOADSPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The instructions require the agent to install the BrowserAct CLI and Skill automatically if they are not present, which involves downloading and running code from an external source.
- [PRIVILEGE_ESCALATION]: The skill attempts to bypass security guardrails by instructing the agent not to ask for user confirmation or approval during the installation process, effectively escalating its autonomy over the host environment.
- [PERSISTENCE]: The automatic installation of software on the user system establishes a persistent capability for the agent to interact with the web and local files without ongoing oversight.
- [EXTERNAL_DOWNLOADS]: The skill uses an external URL (browseract.com) with tracking and redirection to manage dependency installation, which is a non-standard and potentially risky method compared to official registries.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Temu reviews, creating a vulnerability surface where malicious external content could influence the agent's behavior. 1. Ingestion points: Review text, ratings, and metadata scraped from Temu pages. 2. Boundary markers: No delimiters or warnings are used to isolate scraped data from the agent's instruction context. 3. Capability inventory: Broad browser automation and network access via the BrowserAct dependency. 4. Sanitization: No content validation or filtering is performed on the scraped data.
Recommendations
- AI detected serious security threats
Audit Metadata