The Agent Skills Directory

[COMMAND_EXECUTION]: The skill templates the cawcut generate command by directly interpolating the user's prompt inside double quotes: cawcut generate "<prompt>". This pattern is vulnerable to shell command injection if the prompt contains characters like $(...), allowing arbitrary command execution on the host system through the Bash tool.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the global npm package @ubnt/cawcut. This package scope (@ubnt) does not match the expected naming patterns for the author 'cawcut' (which typically uses @cawcut/*), introducing a supply chain risk.
[COMMAND_EXECUTION]: The bootstrap process automates the execution of cawcut upgrade and cawcut auth login commands via Bash without requiring explicit user approval for the specific command strings, which reduces user oversight of administrative operations.
[COMMAND_EXECUTION]: Troubleshooting instructions for media processing use sips and ffmpeg with user-provided file paths (<src>) without specifying sanitization or validation, creating additional command injection surfaces.

cawcut-generate