The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a scaffolding tool that generates new skill files based on user-provided requirements (name, triggers, descriptions, and logic). This creates a structural surface for indirect prompt injection if a user provides malicious instructions intended to be embedded in the generated code. However, the skill includes a mandatory clarification phase where the agent must summarize its understanding and wait for user confirmation before writing any files, acting as a manual validation gate. All generated content is derived from local, trusted templates within the skill's own directory.
[COMMAND_EXECUTION]: The skill instructions direct the agent to use filesystem tools (e.g., mkdir, write_file) to create the directory structure and files for new skills. These operations are performed at the final destination paths specified by the user (either project-local or global configuration directories), which is the primary intended purpose of the tool.
[EXTERNAL_DOWNLOADS]: Reference documentation (references/script-language-guide.md) contains educational examples of how to write scripts that fetch data from external APIs using standard libraries. These are provided as static templates for the user's reference and are not executed by the skill itself during its operation.

skill-creator