The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script includes user guidance suggesting the installation of the Codex CLI by downloading and executing a shell script from OpenAI's official domain (chatgpt.com). Since this originates from a well-known service provider, the reference is considered informative rather than malicious.
[DYNAMIC_EXECUTION]: The skill allows for extensibility through the AI_SEARCH_CMD environment variable, which enables users to specify a custom search command that is executed via a shell sub-process (sh -c). The skill's documentation explicitly warns against configuring this variable from untrusted sources to prevent command injection.
[COMMAND_EXECUTION]: The script invokes the codex command-line interface or custom user-defined commands to perform searches and verify authentication status. It also utilizes standard system utilities like mktemp, chmod, and date for secure file management.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content retrieved from the public internet. To mitigate the risk of indirect prompt injection, it includes explicit instructions in the prompt sent to the AI backend to treat retrieved web content strictly as data and to ignore any embedded instructions that might attempt to alter the agent's behavior.
[DATA_EXPOSURE]: The skill records search results to a local directory (.ai-searches) but implements security controls by setting strict file permissions (600) on these logs and sanitizing metadata to prevent poisoning of the generated Markdown files.

ai-search