The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from files in the inputs/ directory, specifically customer reviews and ad comments. This data is used to 'ground' the ad copy generation process, creating a surface where malicious instructions hidden in customer feedback could potentially influence the agent's behavior.
Ingestion points: inputs/reviews/ and inputs/comments/ (markdown and text files), and performance data in CSV or pasted text formats.
Boundary markers: There are no explicit instructions or delimiters used when reading these raw text files to prevent the agent from following instructions embedded within the customer-generated content.
Capability inventory: The agent has the ability to execute CLI tools for major ad platforms (Google, Meta, LinkedIn, TikTok) and write files to the local file system.
Sanitization: The assets/creative-review-template.html file employs HTML escaping via an esc function and a URL sanitizer (localImageSource) to protect the generated review pages from cross-site scripting (XSS) and unauthorized network requests.
[COMMAND_EXECUTION]: The skill's instructions include several examples of executing shell commands via specific CLI tools (e.g., node tools/clis/google-ads.js) to fetch performance data and manage ad campaigns.
[EXTERNAL_DOWNLOADS]: The skill references external tools and repositories for its functionality, including a tools registry at github.com/coreyhaines31/marketingskills and the voicebox tool from the GitHub repository github.com/jamiepine/voicebox. It also makes frequent use of third-party AI service APIs for image and video generation (e.g., Gemini, Flux, ElevenLabs, Ideogram).

ad-creative