The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection through the processing of untrusted data in .canvas files.
Ingestion points: The workflows for adding nodes and editing canvases in SKILL.md require the agent to read and parse existing JSON files.
Boundary markers: The instructions lack explicit boundary markers or directives to ignore instructions that might be embedded within the text fields of the JSON data.
Capability inventory: The skill grants the agent capabilities to read from and write to the local file system to maintain the canvas files.
Sanitization: There are no instructions for sanitizing or validating the content of the text nodes or the file paths provided in the input data.

json-canvas