ad-creative-brief
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill contains a surface for indirect prompt injection because it processes untrusted data from external sources like customer reviews and competitor ads. However, the risk is negligible as the agent's capabilities are limited to text generation and the ingestion is core to the skill's legitimate function.
- Ingestion points: The skill ingests data from customer reviews, sales-call notes, ad comments, and competitor ad libraries as described in the 'Interview' and 'Evidence-mining methods' sections.
- Boundary markers: No explicit delimiters or instructions to ignore embedded prompts are specified for the ingested data.
- Capability inventory: The agent uses tools for web browsing and skill-to-skill communication (skill chaining) to generate documentation; it does not have access to sensitive file systems or arbitrary code execution.
- Sanitization: The instructions do not define specific sanitization or filtering logic for external content.
- [SAFE]: The skill is well-structured and focuses on performance marketing best practices. It correctly utilizes internal vendor references for related tasks and relies on verified industry methodologies (e.g., Eugene Schwartz's awareness levels). No obfuscation, persistence mechanisms, or unauthorized privilege escalations were detected.
Audit Metadata