google-ads-budget
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from both the Google Ads MCP tool and local workspace markdown files to drive its logic and file-writing operations. The lack of explicit boundary markers for this external data creates a surface for potential manipulation of agent behavior.
- Ingestion points: Output from the
google-ads-mcptool (campaign metrics) and files within theworkspace/ads/directory. - Boundary markers: No specific delimiters or instructions to ignore potential commands within the ingested data are present in the skill instructions.
- Capability inventory: The skill reads local files, writes new files to
workspace/ads/drafts/, and executes GAQL queries via thesearchtool. - Sanitization: There is no evidence of runtime sanitization or validation for the data ingested from the tool or workspace files.
Audit Metadata