ad-creative
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from 'inputs/reviews/' and 'inputs/comments/' to ground the generation of ad creative. This creates a surface where embedded instructions in customer reviews could potentially influence the agent's behavior during the copy generation process.\n
- Ingestion points: The skill reads customer reviews (Trustpilot, G2, Amazon, App Store) from .md and .txt files in the 'inputs/reviews/' directory and campaign comments from 'inputs/comments/' as specified in the 'Grounded Inputs' section of SKILL.md.\n
- Boundary markers: The instructions lack explicit boundary markers or delimiters (such as XML tags or triple quotes) when instructing the agent to incorporate these inputs into the prompt context, relying on natural language instructions to 'cite sources'.\n
- Capability inventory: The skill possesses the capability to execute local CLI tools ('node tools/clis/...') for marketing platform interactions and has write access to the filesystem to generate dated output batches.\n
- Sanitization: The provided 'creative-review-template.html' includes defensive measures such as an 'esc' function for HTML entities and a specific check to block remote URL image sources to prevent tracking beacons.\n- [EXTERNAL_DOWNLOADS]: The skill references an external GitHub repository ('github.com/coreyhaines31/marketingskills') for its 'tools registry' and integration guides for Google, Meta, LinkedIn, and TikTok ads. While these are documented as dependencies for functionality, the repository belongs to an individual account rather than a verified service provider.
Audit Metadata