adcp-media-buy

Pass

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines tools that ingest untrusted data from external sources, creating a surface for indirect prompt injection.
  • Ingestion points: The get_products tool accepts natural language brief inputs. The sync_catalogs and sync_creatives tools accept external URLs for product feeds and creative assets. The list_creative_formats tool fetches data from publisher-hosted .well-known endpoints or the protocol's community mirror.
  • Boundary markers: There are no explicit instructions for the agent to treat these inputs as untrusted data or use delimiters to prevent instruction hijacking.
  • Capability inventory: The skill leverages MCP tools to manage media buys, sync assets, and retrieve performance data.
  • Sanitization: The protocol does not specify sanitization or validation steps for the natural language or remote data content before processing.
  • [EXTERNAL_DOWNLOADS]: The skill references external URLs to fetch creative format specifications and catalog data. Specifically, it uses https://creative.adcontextprotocol.org (the official community mirror for the protocol) and publisher-controlled domains to retrieve configuration files. These are documented protocol behaviors intended for inventory discovery and asset management.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 9, 2026, 10:47 AM