The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by ingesting untrusted user input (command names, descriptions, and implementation steps) and interpolating it directly into generated markdown templates. Ingestion points: User-provided strings during the interactive command creation Q&A (SKILL.md). Boundary markers: None; user input is placed directly into the generated file's instructions without delimiters or warnings. Capability inventory: Uses the Write tool to persist files to the local filesystem and the Bash tool for directory management and tool execution. Sanitization: There is no evidence of validation or escaping for user-provided strings before they are written to command files, which could allow a user to inject instructions that override agent behavior when the generated command is invoked.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform system-level operations, such as directory creation using mkdir -p. It also instructs the agent to execute development tools like git, make, and pytest. While these are expected capabilities for a command creation tool, they involve significant filesystem and system access that requires user oversight of the generated output.

command-creator