The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill contains multiple commands to fetch data from an untrusted external domain.
Evidence: curl -H "Accept: text/markdown" https://fullstackrecipes.com/api/recipes/... is used repeatedly throughout SKILL.md to retrieve instructions.
Risk: The domain fullstackrecipes.com is not a trusted source. Content retrieved at runtime could be modified to include malicious instructions or code that the agent might unknowingly follow.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill is designed to ingest untrusted Markdown content from a remote server, creating a surface for injection.
Ingestion points: curl commands in SKILL.md fetching content from fullstackrecipes.com.
Boundary markers: None present; external content is treated as authoritative instructions.
Capability inventory: The agent is expected to execute setup steps (bash, configuration) based on the downloaded content.
Sanitization: No sanitization or validation of the fetched Markdown is performed.

testing