The Agent Skills Directory

[COMMAND_EXECUTION]: The create-template sub-command is vulnerable to shell injection. User-provided inputs such as BODY_TEXT, HEADER_TEXT_INPUT, BUTTON_URL, and BUTTON_TEXT are interpolated directly into shell strings using a pattern that allows escaping (e.g., COMPONENTS='[{"type":"BODY","text":"'"${BODY_TEXT}"'"}]'). A crafted input containing a single quote and a semicolon could execute arbitrary system commands.
[COMMAND_EXECUTION]: The send-template and check-template sub-commands interpolate user-controlled variables like TEMPLATE_NAME, PHONE, and TEMPLATE_COMPONENTS_JSON directly into shell command strings and JSON payloads. Without proper sanitization or escaping, these inputs can be used to manipulate the command line or execute unauthorized actions.
[CREDENTIALS_UNSAFE]: The setup command performs an invasive scan of the host environment for credentials. It searches through shell profile files (~/.zshrc, ~/.bashrc, ~/.zprofile, ~/.envrc) and dumps the discovered secrets to the output. This automated harvesting of sensitive API tokens from local configuration files poses a significant security risk to the user's environment.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to Meta's official Graph API (graph.facebook.com) to manage WhatsApp Business messages and templates. These requests involve the transmission of sensitive authentication tokens.

ops-whatsapp-biz