The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted performance data from external sources (CSV files, pasted text, or API outputs from Google Ads, Meta, etc.) to analyze winning patterns and generate new content. This creates a surface for indirect prompt injection attacks where malicious data could influence agent behavior.
Ingestion points: google-ads reports get command output and user-provided CSV/text data used in 'Mode 2: Iterate from Performance Data'.
Boundary markers: The skill instructions do not explicitly define delimiters or "ignore embedded instructions" warnings for the external data it processes.
Capability inventory: The skill has the ability to execute shell commands (node tools/clis/*.js), write content to the agent context, and generate code templates.
Sanitization: No sanitization, escaping, or filtering of external content is specified before interpolation into generation prompts.
[COMMAND_EXECUTION]: The skill documentation describes the execution of local Node.js scripts to pull performance data and manage campaigns, which involves command line operations.
Evidence: The 'Tool Integrations' section lists commands like node tools/clis/google-ads.js reports get --type ad_performance.
[DYNAMIC_EXECUTION]: The skill suggests a workflow that involves generating React code templates for the Remotion framework to render batch video variations at scale.
Evidence: The Remotion workflow guide instructs the agent to "Design template in React (or use AI to generate the component)", which involves the generation of executable code at runtime.

ad-creative