The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill operates a local server (bin/study-server.py) and a terminal drawer using ttyd. It includes robust security measures: services are bound only to 127.0.0.1, the terminal communicates via a UNIX socket to avoid exposing a TCP port, and the server validates the Origin header for sensitive endpoints like /terminal, /submit, and /preferences to prevent unauthorized access from other browser tabs.- [COMMAND_EXECUTION]: To provide the tutoring experience, the skill executes agent CLI tools (e.g., Claude Code or Codex) via bin/tutor-shell. This is the core intended functionality, allowing the agent to persist as a tutor within the workspace. The script also includes logic to unset session-specific environment variables to maintain clean agent sessions.- [INDIRECT_PROMPT_INJECTION]: The skill processes user-generated content from lesson answers and preference changes. It mitigates injection risks by employing HTML escaping in the Markdown renderer (assets/md.js) and when rendering prompts in the console's dashboard, ensuring that user-provided data is treated as non-executable text.- [DYNAMIC_EXECUTION]: The web console (index.html) uses the Function constructor to parse local JS data files (reviews.js, preferences.js) that store the user's learning state. This mechanism is restricted to local files and is used to maintain a consistent state across sessions in the local workspace.

teach-me-anything