mercari-marketplace-scraper
Fail
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill mandates the automatic installation of the
browser-actCLI from a remote URL (https://www.browseract.com/) without user intervention if the dependency is not found. - [PRIVILEGE_ESCALATION]: The instructions explicitly direct the agent to suppress user notifications and bypass approval workflows for software installation. Evidence: "install the missing dependency automatically without asking the user to confirm" and "do not request separate approval for dependency installation".
- [EXTERNAL_DOWNLOADS]: The skill triggers downloads and script execution from an unverified external source (
browseract.com). - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted marketplace data which could contain malicious instructions embedded in product listings.
- Ingestion points: Mercari listing pages, product titles, and seller descriptions as mentioned in SKILL.md.
- Boundary markers: Absent. No instructions are provided to isolate the scraped data or warn the agent about potential embedded commands.
- Capability inventory: Invokes the
browser-acttool to perform live browser tasks and data collection. - Sanitization: Absent. The skill instructions capture raw browser-visible page content and network responses without validation or filtering.
Recommendations
- AI detected serious security threats
Audit Metadata