The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines how an agent should create, read, and edit .canvas files, which include user-controlled content in the text attribute of nodes. If an agent processes an untrusted file containing hidden instructions, it could lead to indirect prompt injection.
Ingestion points: The text field of node objects within the JSON structure of .canvas files (as seen in the Text Nodes section of SKILL.md).
Boundary markers: The skill does not define delimiters or specific instructions for the agent to disregard commands embedded within the text attributes.
Capability inventory: The skill enables the agent to perform file system operations (create, edit) on .canvas files.
Sanitization: No sanitization or content validation for the Markdown text provided in node attributes is specified.

json-canvas