human-in-the-loop
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [SAFE]: No malicious patterns or behaviors were detected. The skill is purely informational and serves as a design framework for agent safety.
- [NO_CODE]: The skill is composed of Markdown files containing instructions and examples. It lacks any executable scripts, shell commands, or network-enabled tools.
- [PROMPT_INJECTION]: The skill provides proactive defensive guidance by warning against 'type yes' confirmation gates that are vulnerable to prompt injection from untrusted tool outputs. It recommends out-of-band confirmation methods to mitigate this risk.
- [INDIRECT_PROMPT_INJECTION]: The skill addresses the vulnerability surface of indirect prompt injection by instructing agents to use structured data, explicit policy citations, and human-readable diffs in approval workflows rather than relying on natural language confirmation strings.
Audit Metadata