The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively uses the cmux send-surface command to execute arbitrary shell strings in terminal splits.
[DYNAMIC_EXECUTION]: The skill provides the cmux browser eval command, which allows for the execution of arbitrary JavaScript code within the context of a browser surface.
[DATA_EXFILTRATION]: The skill includes functionality to read potentially sensitive information:
cmux capture-pane --scrollback allows reading the entire history of a terminal session.
cmux browser cookies get and cmux browser state save provide mechanisms to extract session tokens, cookies, and authentication data from browser sessions.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by design.
Ingestion points: Untrusted data enters the agent context through cmux capture-pane (reading terminal outputs), cmux browser snapshot (reading web DOM structures), and cmux browser get text/html (reading web page content).
Boundary markers: Absent. There are no instructions or delimiters provided to help the agent distinguish between data and potential instructions within the captured outputs.
Capability inventory: The agent has significant capabilities including executing shell commands (send-surface), performing browser actions (click, fill, select), and evaluating JavaScript (eval).
Sanitization: Absent. The skill does not demonstrate or require any sanitization, filtering, or validation of the data retrieved from external terminal or browser surfaces before it is processed by the agent.

cmux-terminal-multiplexer