03-performance-eval-global
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns such as prompt injection, obfuscation, or unauthorized data exfiltration were detected. The skill instructions focus on diagnostic methodologies and regional benchmarks.- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection due to its core function of processing external marketing data.
- Ingestion points: User-provided performance metrics, uploaded screenshots, and automated data pulls from Meta, Google, and TikTok MCP servers.
- Boundary markers: The skill relies on structured questions for data gathering but does not explicitly use delimiters or ignore-instructions warnings for user data.
- Capability inventory: Read access to campaign data and gated write access to ad accounts (requiring human approval per Rule 7).
- Sanitization: No explicit sanitization or filtering of external data strings is defined in the instructions.
Audit Metadata