The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides example code in Section 5 ('Backend Integration') to spawn a local shell process (bash or powershell) using the node-pty library. This is the core intended functionality of a terminal emulator, allowing the terminal UI to execute commands on the host machine.
[INDIRECT_PROMPT_INJECTION]: The skill documents an attack surface where untrusted data from a shell process or pasted text could influence terminal behavior.
Ingestion points: Data enters the system via ptyProcess.onData (Section 5) and user keystrokes/pastes via term.onData (Section 6).
Boundary markers: The skill suggests implementing 'Paste Sanitization' (Section 16) to block terminal control characters that could be used for injection.
Capability inventory: The skill uses pty.spawn to create processes and ptyProcess.write to send data to the shell (Section 5).
Sanitization: Section 16 explicitly recommends filtering control characters during paste events and using specific quoting mechanisms ("" and %%) for Windows cmd.exe shims to prevent argument injection.

xtermjs-skill