The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines workflows for creating and editing .canvas files which serve as a persistent ingestion point for untrusted data, including user-provided Markdown text, file paths, and URLs.
Ingestion points: The skill processes .canvas files (JSON structure) containing nodes and edges as described in SKILL.md and references/EXAMPLES.md.
Boundary markers: The instructions do not define specific delimiters or warnings to prevent the agent from following instructions that might be embedded within a text node or group label.
Capability inventory: The skill facilitates file system read and write operations specifically for .canvas files and references local files via the file node type.
Sanitization: There is no mention of sanitizing Markdown content or validating referenced file paths against a restricted directory, which could lead to path traversal if a malicious canvas file is processed.
[EXTERNAL_DOWNLOADS]: The skill references external resources for the JSON Canvas specification and reference implementation.
Provides links to jsoncanvas.org and the github.com/obsidianmd/jsoncanvas repository for technical documentation.

json-canvas