canvas
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Obsidian Canvas (.canvas) files which are JSON-formatted data. These files represent an untrusted input source that could potentially contain instructions intended to influence the agent's behavior.
- Ingestion points: The skill reads the entire canvas file content for inspection and mutation as described in
SKILL.md. - Boundary markers: The instructions require the agent to validate the JSON structure, node types, and dimensions, and to present a preview to the user before applying any changes.
- Capability inventory: The skill can create, update, and replace files within the vault scope using the
claude-obsidian.pytransaction engine, as seen in the 'Preview and apply' section ofSKILL.md. - Sanitization: The skill explicitly instructs the agent to reject absolute paths, directory traversal (
..), home shortcuts, and paths escaping via symlinks for any file references within the canvas.
Audit Metadata