The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it is designed to ingest and process untrusted external data, such as ad performance metrics (provided via CSV, paste, or API tools) and product marketing context files. This data is used by the agent to generate new ad headlines and descriptions.
Ingestion points: Reads .agents/product-marketing-context.md and accepts ad performance data (CTR, conversion rates, etc.) as input for iteration.
Boundary markers: The skill does not define explicit boundary markers or specific instructions to the model to ignore embedded instructions within the ingested context or performance data.
Capability inventory: Across its instructions and references, the skill possesses the capability to execute local CLI tools (e.g., google-ads, meta-ads), perform network operations via curl to external APIs, and write data to local files (e.g., CSV output).
Sanitization: There is no mention of sanitization, validation, or filtering of the external content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to use local command-line tools for interacting with ad platforms. This includes specific commands to execute project-local scripts, such as node tools/clis/google-ads.js, to fetch performance reports and manage campaign assets.
[EXTERNAL_DOWNLOADS]: The documentation contains references to several well-known external AI services and tools. It provides curl command examples for interacting with official APIs (such as Google Gemini and ElevenLabs) and instructions for cloning the voicebox tool from a public repository on GitHub (github.com/jamiepine/voicebox). These references are consistent with the skill's primary purpose of producing ad visuals and audio.

ad-creative