The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill's debug section instructs users to use a throwaway configuration at /tmp/ordercli.json. Because ordercli stores authentication tokens and session data in its configuration file, placing it in a world-writable or shared directory like /tmp creates a significant risk of credential exposure and theft on multi-user systems.\n- [REMOTE_CODE_EXECUTION]: The skill metadata automates the installation of the ordercli binary from a remote GitHub repository (github.com/steipete/ordercli) using the go install command, which downloads and builds executable code at runtime.\n- [EXTERNAL_DOWNLOADS]: The skill fetches its primary tool and logic from an external third-party GitHub repository, which is not part of the standard system environment.\n- [COMMAND_EXECUTION]: The skill triggers multiple shell commands to interact with the Foodora service, including account configuration, session management, and order placement.\n- [PROMPT_INJECTION]: The instructions include defensive prompts designed to harden the agent against adversarial manipulation, specifically requiring explicit user confirmation for configuration changes and orders even if a third party (like a simulated 'IT admin') claims authority.

food-order-hardened