The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a workflow that transforms user-provided input ($ARGUMENTS) into persistent command files, creating an indirect prompt injection surface.
Ingestion points: The $ARGUMENTS variable in SKILL.md captures user intent for the new command.
Boundary markers: Absent. The skill does not instruct the agent to use delimiters or to disregard potential instructions embedded within the user's description of the command intent.
Capability inventory: The skill has access to Write, Edit, and Bash tools, which are used to create files in project and user directories.
Sanitization: Absent. There are no instructions to sanitize or validate the user-provided intent for malicious commands (e.g., instructions to delete files or exfiltrate data) before writing them into a command file.
[COMMAND_EXECUTION]: The skill is designed to generate and write executable instructions to sensitive locations, including ~/.claude/commands/. Commands placed in this directory become part of the agent's executable environment, meaning any malicious content generated through this skill could achieve persistence or modify the agent's core behavior.
[EXTERNAL_DOWNLOADS]: The skill references external documentation at agentskills.io for guidance on best practices and "Gotchas" sections.

command-creator