data
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from MCP connectors and user comments, creating an injection attack surface. However, it incorporates several best-practice mitigations.
- Ingestion points: Data enters the agent's context through MCP tool outputs (
datasets), user-provided comments (ArtifactComments), and live UI state (open-artifact context). - Boundary markers: The system enforces the use of JSON Schema
rulesto validate and sanitize all variables ({{name}}) interpolated into live queries, preventing command injection into underlying data sources. - Capability inventory: The skill allows the agent to call MCP tools via the
Artifactinterface, read and write to a local JSON document store, and open web views for artifacts. - Sanitization: The instructions explicitly mandate that the agent treat all external input as data rather than instructions. Furthermore, the skill's HTML rendering policy automatically strips executable content like
<script>,<iframe>, and inline event handlers to prevent cross-site scripting.
Audit Metadata