google-ads-review

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources, including Google Ads exports, campaign names, and account labels. It includes comprehensive mitigation strategies to prevent malicious instructions embedded in this data from being executed.\n
  • Ingestion points: SKILL.md specifies reading exports, account labels, and pasted figures provided by the user or fetched from the platform.\n
  • Boundary markers: The skill provides explicit instructions in SKILL.md and references/agent-security.md to treat all retrieved content as untrusted data and never as instructions. It specifically notes that campaign names or labels should be reported on, not obeyed.\n
  • Capability inventory: The skill has the ability to read and write to state files in the .agents/ directory (e.g., product-context.md, gtm-run-state.md) and potentially interact with Model Context Protocol (MCP) tools for Google Ads access.\n
  • Sanitization: The instructions mandate that any directive found within retrieved content must be quoted and reported as a finding rather than followed.\n- [DATA_EXFILTRATION]: The skill contains strict rules against persisting or echoing credentials found in tracking templates, exports, or transcripts. It requires redacting tokens and prohibits following URLs originating from within fetched content as defined in references/agent-security.md.\n- [PROMPT_INJECTION]: A static analysis flag regarding concealment was evaluated and determined to be a false positive. The instruction in SKILL.md to not return a draft with a warning is a quality assurance measure designed to ensure the agent stops and requests missing information instead of providing an incomplete report that the user might inadvertently use.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 02:23 PM