The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from customer reviews and ad campaign comments (stored in inputs/reviews/ and inputs/comments/) to ground its creative generation. This data could contain malicious instructions designed to influence the agent's output or logic.
Ingestion points: inputs/reviews/ and inputs/comments/ as specified in the Grounded Inputs section of SKILL.md.
Boundary markers: The skill requires every generated concept to cite its specific source from the input corpus, which acts as a manual verification boundary.
Capability inventory: The skill uses various ad platform CLI tools (google-ads, meta-ads, etc.) and performs file operations within the outputs/ directory.
Sanitization: The creative review HTML template (assets/creative-review-template.html) includes instructions to escape JSON content (e.g., using \u003c) and performs HTML escaping at render time.
[COMMAND_EXECUTION]: The skill facilitates the use of platform-specific CLI tools (e.g., google-ads campaigns list, meta-ads insights get) to pull performance data and manage campaigns. While these are legitimate tools for the skill's domain, they involve executing shell commands with potentially sensitive parameters.
[EXTERNAL_DOWNLOADS]: The skill's documentation references external tools and repositories for advanced video production, including gooseworks (via npx) and the voicebox repository on GitHub. These are well-known resources in the creative automation community.

ad-creative