tmux
Pass
Audited by Gen Agent Trust Hub on Apr 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is designed to send arbitrary keystrokes and commands to tmux panes via the
tmux send-keyscommand. This is a core feature intended to allow the agent to interact with terminal user interfaces (TUIs) and interactive shell sessions. - [PROMPT_INJECTION]: The skill uses
tmux capture-paneto read terminal output, which presents an attack surface for indirect prompt injection. If the terminal displays content from untrusted sources (such as a website or a log file) that contains malicious instructions, the agent might interpret that content as authoritative commands. \n - Ingestion points: Terminal history and current pane content are retrieved in
SKILL.mdandscripts/wait-for-text.sh. \n - Boundary markers: There are no delimiters or explicit instructions provided to the agent to treat captured terminal output as untrusted data. \n
- Capability inventory: The skill allows sending keystrokes, creating or killing sessions, and navigating windows and panes. \n
- Sanitization: Captured output is passed directly to the agent context without validation or filtering.
Audit Metadata