The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest natural language descriptions from users and transform them into executable skill files (SKILL.md).
Ingestion points: The agent reads a "plain-language description" from the user and analyzes existing files in the repository (such as CLAUDE.md and other skills) to generate new content.
Boundary markers: The instructions do not define specific delimiters or "ignore" directives to prevent a malicious user description from influencing the logic of the generated skill.
Capability inventory: The skill uses file-writing capabilities to install new skills and encourages the use of dynamic context injection (shell execution at load time) in the skills it creates.
Sanitization: There are no instructions provided to the agent to filter or sanitize the user input to prevent it from injecting malicious instructions or dangerous shell commands into the resulting skill output.

skill-creator