canvas

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from external Model Context Protocol (MCP) sources, such as Datadog, Sentry, and Slack, to generate visual artifacts. This creates a surface for indirect prompt injection where malicious content in those external sources could influence the agent's behavior during artifact creation.
  • Ingestion points: External tool data retrieved via MCP (SKILL.md).
  • Boundary markers: The instructions do not mandate the use of delimiters or boundary markers to isolate untrusted data.
  • Capability inventory: The agent utilizes file system write access to create .canvas.tsx files (SKILL.md).
  • Sanitization: The skill does not explicitly instruct the agent to sanitize or escape external content before rendering it in the UI.
  • [DATA_EXFILTRATION]: The skill directs the agent to interact with files inside a hidden application configuration directory in the user's home folder.
  • Evidence: The skill defines the primary storage path as /Users/<user>/.cursor/projects/<workspace>/canvases/ (SKILL.md). While this is the intended location for this specific IDE feature, it involves interaction with sensitive home directory subfolders.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 10:36 AM
Security Audit — agent-trust-hub — canvas