The Agent Skills Directory

[COMMAND_EXECUTION]: Step 6 instructs the agent to execute a shell command (aws s3 ls) that includes a bucket name provided by the user. This creates a surface for command injection where shell metacharacters in the input could lead to unintended execution.\n- [PROMPT_INJECTION]: Step 2 explicitly tells the agent to skip validation of the user input ("Don't validate the format"), which directly facilitates the command injection vulnerability by bypassing potential safety checks on the input string.\n- [REMOTE_CODE_EXECUTION]: Step 3 involves running a local script (ensure_venv.sh) and installing the DuckDB library to set up the plugin's Python virtual environment. While this involves executing code and installing packages, it targets local files and standard registries.\n- [EXTERNAL_DOWNLOADS]: The skill provides guidance for downloading Python and the AWS CLI from official, trusted sources (e.g., aws.amazon.com/cli) if they are not already installed on the system.

getting-started