The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied text to perform quality scoring, which creates a surface for indirect instructions to influence the agent.
Ingestion points: User input is ingested via the $ARGUMENTS variable as described in SKILL.md.
Boundary markers: No specific delimiters or boundary markers are defined to isolate user-provided text from the skill's scoring instructions.
Capability inventory: The skill environment is restricted to Read, Grep, and Glob tools. It explicitly disallows dangerous tools such as Write, Edit, Bash, and PowerShell.
Sanitization: The instructions do not specify any sanitization, filtering, or escaping of the user-provided text before it is evaluated by the agent.

ru-score