The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data, such as learning topics and answers to questions, and stores them in persistent files including session.md and learner-profile.md. These files are later read by the agent to restore session state or adapt its teaching style, creating an opportunity for malicious user input to influence future agent logic.
Ingestion points: User-provided learning topics and answers to diagnostic and Socratic questions (SKILL.md).
Boundary markers: Not present; user data is directly interpolated into markdown tables and text logs without delimitation or safety warnings.
Capability inventory: File system read/write, shell command execution (open), and interactive question tools (AskUserQuestion).
Sanitization: The skill instructions do not specify any validation, filtering, or escaping of user-provided content before it is committed to persistent storage.
[COMMAND_EXECUTION]: The skill instructions direct the agent to use the shell open command to launch the system's default web browser for viewing generated learning roadmaps and visual aids.
[EXTERNAL_DOWNLOADS]: The HTML templates for Excalidraw diagrams and visual components fetch external libraries, such as React and Excalidraw, from the esm.sh CDN at runtime when the files are opened in a browser.
[DYNAMIC_EXECUTION]: The skill dynamically generates HTML and JavaScript files (roadmap.html, summary.html, visuals/*.html) based on the current session state and user input, which are then executed within the local browser environment.

sigma