The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses a stateful workspace where the agent's behavior is guided by reading files such as MISSION.md, RESOURCES.md, and learning-records/. This creates an attack surface where data retrieved from external resources or previous user interactions could influence the agent's logic in future sessions.
Ingestion points: The agent is instructed to populate RESOURCES.md with external citations and read learning-records/ to determine the "zone of proximal development."
Boundary markers: None are specified in the provided templates to isolate untrusted content from the agent's instructions.
Capability inventory: The agent can write to the local file system and execute CLI commands to open files.
Sanitization: There are no specific instructions for the agent to sanitize or validate external content before storing it in the workspace files.
[DYNAMIC_EXECUTION]: The skill's primary function is to generate interactive HTML lessons and reusable assets (like quiz widgets and simulators) in the ./assets/ and ./lessons/ directories. This involves the generation of executable client-side code (HTML/JavaScript) by the agent, which is then served locally to the user.
[COMMAND_EXECUTION]: The instructions direct the agent to "open the lesson file for the user by running a CLI command" after authoring it. This pattern involves the execution of shell commands based on file paths generated during the session.

teach