coupang-shopping
Fail
Audited by Gen Agent Trust Hub on Mar 14, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill instructions require the agent to store highly sensitive credentials in a plain-text file on the local disk.
- Evidence: Instructions direct the creation of
~/.coupang-session/credentials.jsoncontainingemail,password, andpaymentPin. - [COMMAND_EXECUTION]: User-provided product names and search terms are directly interpolated into shell commands, which can lead to command injection if a user provides malicious input.
- Evidence: The prompt templates for the sub-agent include
npx coupang-cli search "[사용자 요청]"andnpx coupang-cli order-now "상품명"where variables are enclosed in quotes but not sanitized for shell metacharacters. - [EXTERNAL_DOWNLOADS]: The skill automatically installs external software packages and browser binaries during initialization.
- Evidence: Use of
npm install -g coupang-cliandnpx playwright install firefoxto set up the environment. - [DATA_EXFILTRATION]: The skill captures and processes sensitive screen data, including keypad images that contain numerical mapping for secure PIN entry.
- Evidence: The "주문 시 키패드 처리 절차" involves taking 10 screenshots of the keypad (
pad-key-0.pngto9.png), reading them, and creating a mapping file.
Recommendations
- AI detected serious security threats
Audit Metadata