ad-creative
Warn
Audited by Gen Agent Trust Hub on May 6, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The documentation in
references/generative-tools.mdprovides instructions to download code from an untrusted source (jamiepine/voicebox) usinggit clone. - [REMOTE_CODE_EXECUTION]: The workflow for video production described in
references/generative-tools.mdencourages the dynamic generation of React components at runtime, which are then executed by the Remotion tool. - [COMMAND_EXECUTION]: The skill uses shell commands and local scripts (e.g.,
node tools/clis/google-ads.js) to interact with advertising platforms and perform media processing withffmpegas detailed inSKILL.md. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of external performance metrics.
- Ingestion points: Performance data is ingested from CSV files, text pastes, or API outputs as described in
SKILL.md(Mode 2). - Boundary markers: There are no instructions or patterns provided to use delimiters or ignore embedded instructions within the untrusted performance data.
- Capability inventory: The agent has access to powerful advertising platform APIs (Google, Meta, LinkedIn, TikTok) and local shell execution.
- Sanitization: No data validation or sanitization steps are defined to filter the external content before it is processed by the agent.
Audit Metadata