The Agent Skills Directory

[SAFE]: The skill consists entirely of natural language instructions for the AI to adopt a critical persona. It does not include any scripts, dependencies, or tool invocations that could pose a security risk.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input such as files, snippets, or plans for review. While this creates an entry point for indirect prompt injection, the risk is negligible as the skill lacks any capabilities to perform network exfiltration, file modifications, or command execution. The instructions specifically mandate that the AI stop at identifying findings and avoid proposing fixes, which restricts the potential for malicious instructions in the input to influence the system beyond the immediate conversation context.

roast