The Agent Skills Directory

[SAFE]: The skill uses a structured 'design tree' approach to interview users. While it mentions using sub-agents to query the filesystem or tools for facts, this is described as a functional requirement for gathering context and does not include specific malicious commands or hardcoded sensitive paths.
[INDIRECT_PROMPT_INJECTION]: There is a potential attack surface where sub-agents could ingest untrusted data from the filesystem or tools. However, the instructions focus on logical decision-making and factual retrieval, and no specific exploits or unsafe interpolation patterns were identified.

grilling