skills/gpu-cli/skills/tmux-cli-test/Gen Agent Trust Hub

tmux-cli-test

Warn

Audited by Gen Agent Trust Hub on Feb 23, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION] (MEDIUM): The skill provides functions like tmux_start and _docker_tmux that execute arbitrary shell commands provided as strings. While necessary for testing, this capability can be abused to execute malicious code if the agent is manipulated into testing a crafted command string.
  • Evidence: tmux_start <session> <cmd> in SKILL.md and docker exec "$TMUX_DOCKER_CONTAINER" tmux "$@" in tmux_docker_helpers.sh.
  • [PROMPT_INJECTION] (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8). It captures terminal output (tmux_capture, docker_tmux_capture) and feeds it back to the agent. If the application being tested displays attacker-controlled data (e.g., from a web page, file, or database), that data could contain instructions that influence the agent's next actions.
  • Ingestion points: tmux_capture and docker_tmux_capture functions.
  • Boundary markers: None. The skill reads raw terminal output without sanitization or explicit instruction boundaries.
  • Capability inventory: The skill can run shell commands, interact with Docker, and send keypresses.
  • Sanitization: No escaping or validation of terminal output is performed before it is processed by the agent.
  • [EXTERNAL_DOWNLOADS] (SAFE): The skill references an internal helper script tmux_helpers.sh via a relative path. While this file was not provided for analysis, it is treated as a local dependency of the skill rather than an external/remote download.
  • Reference: source .claude/skills/tmux-cli-test/scripts/tmux_helpers.sh.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 23, 2026, 05:34 AM