The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input via the $ARGUMENTS variable, which is directly interpolated into the task description. This description is used to guide agent selection and is passed to subagents as their core mission. Without delimiters or sanitization, a malicious task could influence or override the instructions of the coordinator or the subagents.\n
Ingestion points: The $ARGUMENTS placeholder in SKILL.md receives raw user input.\n
Boundary markers: There are no protective delimiters (e.g., XML tags or triple quotes) or "ignore" instructions surrounding the user input.\n
Capability inventory: The skill utilizes the Task tool to spawn subagents, reads agent definitions from .claude/agents/, and writes output to the docs/ directory.\n
Sanitization: No validation or filtering is applied to the task description before it is processed.

team