The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied descriptions and names to generate new SKILL.md files and directory structures.
Ingestion points: User input provided via the argument-hint (<skill-name> <description>) and natural language triggers.
Boundary markers: The skill uses a defined YAML frontmatter and Markdown template ('SKILL.md Format Reference') to structure the output, which helps mitigate accidental injection but does not prevent adversarial input.
Capability inventory: The skill utilizes Bash (for mkdir, ls, and grep) and Write (to create the new skill file) tools.
Sanitization: The instructions direct the agent to 'extract kebab-case identifier' for the skill name, which acts as a basic validation step, but no explicit sanitization of the description content is performed before writing it to the file.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform directory management (mkdir -p) and file searching (ls | grep). While these are standard administrative tasks for a scaffolding tool, they are performed using parameters derived from user input.

skill-creator