The Agent Skills Directory

[PROMPT_INJECTION]: The skill is highly vulnerable to Indirect Prompt Injection due to its core 'grounding' mechanic. It explicitly instructs the agent to ingest large amounts of untrusted data from external sources, including customer reviews (from sites like Trustpilot, G2, Amazon, and App Store) and ad comments.
Ingestion Points: inputs/reviews/ (.md/.txt), inputs/comments/, and performance data (CSV or API outputs).
Boundary Markers: There are no specific boundary markers or 'ignore' instructions provided for the raw text ingestion in SKILL.md or references/hook-system.md.
Capability Inventory: The skill leverages powerful tools for google-ads, meta-ads, linkedin-ads, and tiktok-ads, which can read and write campaign data. Malicious instructions embedded in a processed review could potentially trick the agent into misconfiguring campaigns or leaking performance data.
Sanitization: The references/creative-review-page.md file demonstrates security awareness by instructing agents to escape < as \u003c when generating JSON for the HTML template, but this protection is specific to the review page and does not cover the primary generation logic.
[DATA_EXFILTRATION]: The skill provides instructions for the agent to use specialized tools to pull sensitive performance data from Google, Meta, LinkedIn, and TikTok ad accounts. While the skill doesn't contain exfiltration code, the combination of access to financial/marketing data and the ingestion of untrusted external content (Category 8) creates a risk that the data could be exfiltrated via an injection attack.
[EXTERNAL_DOWNLOADS]: The skill references several third-party repositories and packages for specialized ad production. It suggests users install tools via npx gooseworks install --all and clone external repositories like jamiepine/voicebox for voice generation. These are presented as part of the recommended developer workflow for advanced ad formats.
[COMMAND_EXECUTION]: The skill documents the use of local CLI tools for platform management, such as node tools/clis/google-ads.js, to fetch reports and manage campaigns. It also suggests using ffmpeg via the command line for media processing in its video generation pipelines.

ad-creative