ad-creative
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to ingest and iterate on performance data from external sources (CSV, API output, or text). This provides a vector for indirect prompt injection as the agent may process malicious instructions embedded in the data.
- Ingestion points: External performance data (SKILL.md, Mode 2; evals.json, ID 3).
- Boundary markers: Absent. No delimiters or sanitization directives are provided.
- Capability inventory: Access to ad platform tools (Google, Meta, LinkedIn, TikTok) that can create and modify campaigns.
- Sanitization: Absent.
- [EXTERNAL_DOWNLOADS]: The skill's reference documentation (references/generative-tools.md) encourages users to download and run code from external sources like github.com/jamiepine/voicebox.git and use npx create-video@latest.
- [COMMAND_EXECUTION]: Workflow examples in SKILL.md reference local scripts (e.g., tools/clis/google-ads.js) that are not provided in the skill package, indicating a dependency on external code execution context.
Audit Metadata