The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to perform filesystem operations including creating directories and writing new files to the user's project environment based on provided template names and paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied strings and project-local template files to generate new documents, creating an attack surface for indirect prompt injection or path traversal if inputs are not validated by the platform.
Ingestion points: User-provided Title, Slug, Audience, and Output directory along with project-local Typst template files (templates/sop/*.typ).
Boundary markers: Absent. There are no instructions for the agent to use delimiters or ignore embedded instructions within the ingested templates.
Capability inventory: The agent uses filesystem read, directory creation, and filesystem write capabilities to perform the scaffolding task.
Sanitization: The instructions do not specify a requirement to sanitize user-provided paths or content before writing to the local disk.

new-sop