The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data provided through $ARGUMENTS or retrieved from recent context (Ingestion point) to perform linguistic analysis. The instructions do not specify the use of boundary markers or delimiters for the ingested text (Boundary markers). The skill's capabilities are limited to read-only operations using allowed-tools like Read and Grep, while shell and write tools are explicitly disabled via disallowed-tools (Capability inventory). The skill does not mention specific sanitization or escaping mechanisms for the input text (Sanitization).
[SAFE]: The skill proactively restricts its execution environment by disabling dangerous tools such as Bash, PowerShell, and Write, significantly reducing the attack surface.
[SAFE]: Resource access is confined to a local directory of reference files, with explicit instructions to use file-system tools instead of shell commands for discovery.

ru-check