The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The routing logic evaluates natural language indicators to classify tasks, which creates an attack surface where maliciously crafted input could influence agent delegation.
Ingestion points: Task strings evaluated in the 'classify_task' and 'task_type_routing' sections.
Boundary markers: Absent; there are no explicit instructions to ignore nested commands or use delimiters for untrusted task data.
Capability inventory: This is a no-code skill with no direct execution capabilities (no subprocess, file-write, or network operations).
Sanitization: No input sanitization or validation logic is defined within the routing rules.

delegation-router