The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted HTML and text data from external landing pages via the WebFetch tool.
Ingestion points: Fetched landing page content from user-supplied or discovered URLs in Phase 2.
Boundary markers: Not explicitly defined; the instructions do not specify delimiters to isolate external content from the agent's instructions.
Capability inventory: The skill utilizes runScript for Google Ads Query Language (GAQL) operations and WebFetch for external network requests.
Sanitization: No specific filtering or escaping is mentioned for the ingested HTML content before processing.
[EXTERNAL_DOWNLOADS]: The skill fetches diagnostic data from the official Google PageSpeed Insights API (googleapis.com). This uses a well-known service for its intended analytical purpose and is considered safe.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads and writes to local files including business-context.json and landing-page-history.json. These operations are transparently described as necessary for retrieving business goals and maintaining an audit trail.

google-ads-landing