agency-persona-walkthrough-specialist
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [SAFE]: The skill consists exclusively of Markdown documentation, psychological frameworks, and reporting templates. It lacks any functional code, shell scripts, or automation logic that could pose a security risk during execution.
- [DATA_EXPOSURE]: There are no references to sensitive file paths (e.g., .ssh, .aws), hardcoded secrets, or environment variables. The skill does not include any tools or instructions for network-based data exfiltration.
- [REMOTE_CODE_EXECUTION]: The analysis found no evidence of external package installation (pip, npm) or remote script execution (curl/wget piped to a shell). The workflow is entirely based on qualitative analysis and text generation.
- [OBFUSCATION]: The instructions are provided in plain text. No Base64 encoding, zero-width characters, homoglyphs, or other techniques to hide malicious intent were detected.
- [PROMPT_INJECTION]: The skill provides structured guidance for persona simulation without attempting to override the base agent's safety constraints or system instructions. It focuses on maintaining analytical rigor and human-like monologues within a UX context.
- [INDIRECT_PROMPT_INJECTION]: While the skill involves analyzing external web pages (via screenshots), it does not possess any capabilities (such as file-writing, code execution, or network requests) that could be exploited by malicious content embedded in the analyzed data. The attack surface is effectively zero.
Audit Metadata