The Agent Skills Directory

[METADATA_POISONING]: The skill's author field in the metadata is set to 'google-ads-api-team', which conflicts with the actual skill author ('samaeldemiurgos2-cyber'). This misrepresentation of origin is a deceptive practice.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process API release notes and compatibility requirements from external URLs to dynamically update code templates. This behavior creates a theoretical surface where malicious instructions on those pages could influence the agent's output.
Ingestion points: Sections A and B in SKILL.md direct the agent to read external Google documentation and GitHub repository metadata.
Boundary markers: The skill does not provide instructions for the agent to ignore or delimit instructions that might be embedded in the fetched external data.
Capability inventory: The skill generates code and configuration templates across multiple languages but does not instruct the agent to execute them directly.
Sanitization: The agent is instructed to perform basic string replacement for version placeholders based on the retrieved data.
[EXTERNAL_DOWNLOADS]: The skill retrieves API version information and language support details from official Google domains and repository paths. These are well-known services and trusted sources for API metadata.

google-ads-api-quickstart