tmux
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides tools for driving tmux terminal panes. It uses a safety denylist in
scripts/safe-send.shto block command sending to sensitive interactive applications likevim,ssh, andtop, ensuring the agent only interacts with shells or REPLs.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes terminal history which could contain untrusted content. It mitigates this risk inscripts/run-in-pane.shby using unique, randomized sentinel tokens to delimit command output.\n - Ingestion points: Pane output via
capture-pane(documented inSKILL.mdand used in scripts).\n - Boundary markers: Randomized start/end tokens in
run-in-pane.sh.\n - Capability inventory: Shell command execution and keystroke simulation.\n
- Sanitization: Denylist of interactive programs in
safe-send.sh.
Audit Metadata