The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to review external, learner-provided content, which introduces a potential attack surface where malicious instructions could be embedded in student submissions to influence the agent's behavior.
Ingestion points: The agent is instructed to read content from task.md, assignment.html, and various learner-created artifacts located in the assignments/ directory (referenced in SKILL.md).
Boundary markers: The instructions do not specify the use of clear delimiters or explicit "ignore embedded instructions" warnings when the agent processes the content of these external artifacts.
Capability inventory: The skill allows for the reading and writing of workspace files (AGENTS.md, reviews/) and suggests that the agent perform "execution" and "tests" on the submitted artifacts to verify them.
Sanitization: There are no instructions for sanitizing, escaping, or filtering the content of external artifacts before they are read or processed by the agent.

editor