The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a CLI tool named meta to interact with the Meta Ads API. It uses this tool to fetch ad performance insights and creative metadata.
Evidence: Usage of meta ads insights get, meta ads ad get, and meta ads creative get in Step 1 of the skill process.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data from the Meta Ads API and visual text extracted from user-uploaded images, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Ad performance data from the Meta API (via meta CLI), creative metadata, and text identified in images via vision capabilities.
Boundary markers: None identified. The skill does not include specific delimiting instructions to treat the analyzed ad data as untrusted content.
Capability inventory: The agent has the ability to execute shell commands (the meta CLI) and write files to the local workspace/ directory.
Sanitization: None identified. Data fetched from the external API or extracted from images is interpolated into the generation process without validation or filtering for prompt injection patterns.

ad-copy-generator