diverse-content-gen
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [Metadata Poisoning] (SAFE): The skill metadata accurately reflects its purpose. No deceptive instructions or malicious patterns found in the frontmatter.
- [Prompt Injection] (SAFE): The provided prompt templates use natural instructional language and focus on structured output constraints (JSON) rather than bypassing safety filters or overriding core agent behavior.
- [Data Exposure & Exfiltration] (SAFE): No hardcoded credentials, sensitive file paths, or network operations were found in the documentation or templates.
- [Indirect Prompt Injection] (SAFE): The skill describes workflows for processing external data (e.g., from files like 'products.txt'). 1. Ingestion points: references/tool-integration.md describes reading content from user-specified files. 2. Boundary markers: Templates use structured JSON blocks and clear 'USER REQUEST' headers to delimit external input. 3. Capability inventory: references/tool-integration.md describes using read_file, write_file, and edit_file tools. 4. Sanitization: Not explicitly implemented in pseudo-code, but the use of structured output formats (JSON) reduces the risk of schema confusion.
Audit Metadata