The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's design involves ingesting untrusted user content and interpolating it into a markdown brief file, which is a potential surface for indirect prompt injection.
Ingestion points: The Required inputs section in SKILL.md specifies that the skill accepts project names, descriptions, and reference URLs from the user.
Boundary markers: The skill does not provide delimiters or instructions to treat user input as data only, lacking protection against embedded instructions.
Capability inventory: The workflow in SKILL.md includes a file-write capability to generate a BRIEF.md file at the project root.
Sanitization: No mechanisms for escaping or filtering user input are described before the data is written to the persistent output file.

creative-direction