The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes an execute_typescript tool to handle large-scale grading tasks. This involves generating and running TypeScript code strings at runtime, including importing local modules such as ./canvas/grading/bulkGrade.js. Runtime execution of dynamically constructed code is a significant security risk, as it allows for potential exploitation of the agent's local environment.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes data from student submissions, which are untrusted external sources. These submissions could contain instructions designed to deceive the LLM and manipulate grading outcomes.
Ingestion points: Untrusted student content is ingested via the list_submissions tool in SKILL.md.
Boundary markers: There are no explicit instructions or delimiters used to isolate student content or warn the agent to ignore embedded instructions within that content.
Capability inventory: The skill possesses high-privilege capabilities including grade_with_rubric, bulk_grade_submissions, and execute_typescript, enabling it to modify student records and execute code.
Sanitization: The skill documentation does not specify any sanitization, validation, or filtering of the student submission data before it is analyzed by the AI or used within the code execution environment.

canvas-bulk-grading