blueprint-mcp-diagram-generation
Pass
Audited by Gen Agent Trust Hub on May 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface by ingesting untrusted data and interpolating it into prompts used for diagram generation.
- Ingestion points: Data is pulled from various external sources including local files (
read_file), GitHub repositories (github_list_files), HubSpot deals (hubspot_get_deal), and Google Drive documents (gdrive_read_file). - Capability inventory: The skill possesses network capabilities to send these prompts to the Arcade API via the
start_diagram_jobtool. - Boundary markers: The instructions do not define clear delimiters or "ignore instructions" markers when embedding external content into the diagram prompt templates.
- Sanitization: No sanitization, escaping, or validation of the ingested content is described before it is processed by the LLM.
- [SAFE]: The skill interacts with well-known services (Arcade AI, Google AI Studio) and provides standard security guidance, such as using a CLI tool to store secrets rather than hardcoding them in files.
Audit Metadata