The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources, making it vulnerable to indirect prompt injection attacks.
Ingestion points: The skill reads customer reviews from inputs/reviews/, ad campaign comments from inputs/comments/, and external performance data via tool outputs.
Boundary markers: Instructions lack explicit delimiters or "ignore embedded instructions" warnings when processing the contents of these external files.
Capability inventory: The skill has broad capabilities including filesystem access, network communication with multiple APIs (Gemini, Meta, Google Ads, ElevenLabs), and shell command execution.
Sanitization: There is no provided mechanism for sanitizing or filtering instructions that might be embedded within the customer reviews or comments before they are interpolated into prompts for creative generation.
[EXTERNAL_DOWNLOADS]: The skill guides the agent to download and execute code from third-party repositories and package registries that are not included in the trusted vendor list.
Recommends installing the gooseworks pipeline using npx gooseworks install --all for rendering iMessage-style ads.
Instructions include cloning the voicebox repository from GitHub (jamiepine/voicebox) and running local build scripts (make setup).
While these tools are relevant to the skill's purpose, they involve running code from unverifiable external sources.
[COMMAND_EXECUTION]: The skill frequently instructs the agent to invoke shell commands and local scripts to process data and generate media.
Executes local Node.js scripts for platform integrations: node tools/clis/google-ads.js.
Uses ffmpeg for complex audio-visual stitching and normalization.
Relies on npx for rendering video variations via remotion.

ad-creative