tmux-agent-comms
Pass
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses tmux and python3 subprocess calls to manage agent sessions. This is implemented using standard best practices, including list-based arguments for subprocess.run to prevent shell injection and explicit checks for tmux availability.
- [PROMPT_INJECTION]: The skill processes data from external agents, which constitutes an indirect prompt injection surface.
- Ingestion points: Data is captured from tmux panes via capture-pane (executed in scripts/wait_for_idle.py and scripts/preflight_send.py).
- Boundary markers: The system uses a 'proof cycle' with unique, fresh completion markers (e.g., TAC_DONE_[suffix]) and baseline captures to distinguish new agent output from prompt echo or previous turns.
- Capability inventory: The orchestrator can run shell commands (via tmux), manage local temporary files, and execute Python scripts.
- Sanitization: The skill recommends stripping TUI chrome from captured panes and warns the agent to relay only substantive content, though it does not perform automated semantic sanitization of agent replies.
Audit Metadata