canvas

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses a local script, claude-obsidian.py, which is part of the product installation, to handle vault transactions. This allows for controlled inspection and application of changes to JSON Canvas files through command-line arguments.
  • [DYNAMIC_EXECUTION]: The skill provides instructions for the agent to use python3 -c for extracting image metadata using the Pillow library and the identify utility. These shell-based executions are scoped to files within the user's vault for the purpose of node positioning.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by reading existing .canvas JSON files (Ingestion Point: SKILL.md). While it does not utilize explicit prompt delimiters, it provides a structured validation workflow (Boundary Marker: SKILL.md) and enforces vault-relative path restrictions to prevent traversal (Sanitization: SKILL.md). The capability inventory includes transaction management tools and image metadata extraction (Capability Inventory: SKILL.md, references/canvas-spec.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 05:50 PM
Security Audit — agent-trust-hub — canvas