chatgpt-ads-launch
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data from advertiser ad exports (XLSX/CSV) and website domains to generate campaign workbooks.
- Ingestion points: Ad export files and landing page URLs provided during the prompt sequence (SKILL.md, Prompt 1).
- Boundary markers: The instructions include clear constraints for the agent, such as "review only these sources" and explicit prohibitions against requesting credentials or private customer data.
- Capability inventory: The skill focuses on text generation and workbook population. It does not possess autonomous capabilities for sensitive file access or unauthorized network operations.
- Sanitization: The workflow incorporates multiple validation steps and a mandatory human review policy ("A qualified human on the team must review every ad... before upload") to mitigate risks from potentially malicious or incorrect generated content.
Audit Metadata