monitor-with-tmux
Warn
Audited by Gen Agent Trust Hub on May 11, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill heavily utilizes the
tmuxCLI to create sessions (new-session), send arbitrary keystrokes (send-keys), and capture pane content (capture-pane). These operations allow the agent to execute shell commands and interact with terminal environments based on user-provided instructions. - [REMOTE_CODE_EXECUTION]: The skill instructs the agent to write a Python script (
wait_tmux.py) to the local file system at runtime and subsequently execute it using the Python interpreter. This dynamic code generation and execution pattern is a common technique for bypassing static analysis of executable content. - [DATA_EXFILTRATION]: The monitoring functionality relies on
tmux capture-paneto read terminal output. If sensitive information—such as credentials, API keys, or private data—is printed to the terminal during the monitoring process, it is captured into the agent's context, leading to potential data exposure. - [PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by taking natural language instructions (e.g., 'login', 'ssh', 'goto') and mapping them to executable shell commands inside a tmux session. There are no boundary markers or sanitization steps to prevent malicious instructions from being executed if the user-provided data is manipulated.
Audit Metadata