The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file scripts/subscribr-login.js contains hardcoded authentication credentials, including an email address ('jasonnoahchoi@gmail.com') and a password ('Latch7-Giggling-Flop').\n- [COMMAND_EXECUTION]: The installation script scripts/install-deps.sh utilizes sudo to install system-level packages across multiple Linux distributions (Ubuntu, Fedora, Arch, etc.). Additionally, scripts/screenshot.js uses execSync to execute ImageMagick commands where file paths are constructed from command-line arguments, which may lead to command injection if input is not properly sanitized.\n- [REMOTE_CODE_EXECUTION]: The script scripts/evaluate.js implements an eval() call within the browser context to execute arbitrary JavaScript provided via the --script argument.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Ingestion points include scripts/evaluate.js (executing JS in page), scripts/network.js (monitoring traffic), and scripts/console.js (monitoring logs). Capability inventory includes subprocess execution (scripts/screenshot.js), file system writes (scripts/network.js), and arbitrary JS evaluation. No boundary markers or sanitization routines are present to distinguish between web data and agent instructions.

chrome-devtools