json-canvas
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references the official JSON Canvas specification and its associated GitHub repository for guidance on file structure.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from existing
.canvasfiles, creating a surface for indirect instruction injection.\n - Ingestion points: The workflows in
SKILL.mdfor adding nodes to an existing canvas or editing an existing canvas involve reading and parsing external.canvasfiles.\n - Boundary markers (absent): No specific delimiters or instructions for the agent to ignore embedded commands within fields like
textorlabelare defined.\n - Capability inventory: The skill enables the agent to read and write local
.canvasfiles (identified inSKILL.md).\n - Sanitization (absent): Validation instructions are focused on JSON structure and unique identifiers, but do not provide rules for sanitizing or escaping the content of textual data fields.
Audit Metadata