The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a web terminal component that processes data from external sources (such as a WebSocket PTY stream), which creates a surface for potential indirect prompt injection if an AI agent reads the terminal output.
Ingestion points: The wsUrl parameter in createTerminal and useTerminal, as well as the onData callback in SKILL.md.
Boundary markers: None identified; the system renders raw VT100/xterm sequences directly into the DOM.
Capability inventory: The example backend spawns a shell process (bash) and supports bidirectional communication through the terminal interface.
Sanitization: The terminal uses a Zig-based WebAssembly core for sequence parsing, though it does not implement specific delimiters to prevent an AI from interpreting rendered text as instructions.
[COMMAND_EXECUTION]: The skill provides a Node.js backend implementation using node-pty to spawn local shell processes (e.g., bash), allowing the user to execute system commands through the browser-based terminal interface.
[EXTERNAL_DOWNLOADS]: The development setup section instructions include cloning the project source code from Vercel Labs' official GitHub repository.
[DYNAMIC_EXECUTION]: The terminal core is implemented as a WebAssembly (WASM) binary (wterm.wasm) which is loaded and instantiated at runtime to handle terminal parsing logic.

wterm-web-terminal