The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting external data sources, including .agents/product-marketing-context.md and performance metrics provided via CSV, paste, or API. This data is used to inform creative generation. The instructions lack explicit boundary markers or sanitization requirements for this untrusted input, which is a potential surface for indirect prompt injection.
Ingestion points: SKILL.md (Performance data analysis Mode 2), marketing context loading from local files.
Boundary markers: Absent from prompts.
Capability inventory: Creative generation, platform-specific drafting, and suggesting CLI operations for ad platforms.
Sanitization: No explicit filtering or validation of external content mentioned.
[EXTERNAL_DOWNLOADS]: The references/generative-tools.md file contains instructions for setting up third-party tools, including fetching projects from GitHub (e.g., git clone https://github.com/jamiepine/voicebox.git) and using package executors like npx create-video@latest. These are documented as installation procedures for external dependencies.
[COMMAND_EXECUTION]: The skill includes illustrative bash examples for interacting with advertising platform CLIs (such as google-ads, meta-ads, and tiktok-ads) and for media processing tasks using ffmpeg and remotion. These snippets serve as workflow examples for integrating the skill's output with external systems.

ad-creative