using-tmux-agent-tools
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill orchestrates data flow between a commander and multiple worker agents, which naturally creates a surface for indirect prompt injection. The instructions proactively mitigate this by establishing a boundary marker, explicitly informing the agent that worker output should be treated as data and not as instructions. Evidence of the injection surface includes ingestion points like the 'brief' files and result output; boundary markers are present in the supervisor instructions; and capabilities are limited to authorized tmux-agent tools.
- [DYNAMIC_EXECUTION]: The skill is designed to manage the lifecycle and execution of external agent processes. This behavior is the intended primary functionality and is conducted via specific, documented MCP tools and CLI wrappers (e.g., 'agent-tmux', 'mcp__tmux-agent__assign') within a supervised framework.
- [DATA_EXPOSURE]: The skill references local configuration and state files (e.g., '~/.config/agent-tmux/profiles/', 'result.json') which are standard for the operation of the referenced infrastructure. These interactions are confined to task management and do not involve unauthorized access to sensitive user data or system credentials.
Audit Metadata