Skills
skills/davidondrej/skills/delegating-to-agents/Gen Agent Trust Hub

delegating-to-agents

Warn

Audited by Gen Agent Trust Hub on Jul 18, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides detailed instructions for using cmux and tmux to send raw shell commands and keystrokes to other agent processes.
  • [COMMAND_EXECUTION]: The instructions explicitly recommend using the --permission-mode bypassPermissions flag for Claude Code and the --yolo flag for Codex. These flags are designed to suppress user confirmation prompts and safety checks, allowing the agents to execute potentially destructive actions autonomously.
  • [COMMAND_EXECUTION]: Includes instructions for establishing SSH connections to remote VPS instances and launching agents with high-autonomy flags directly on the remote host.
  • [PROMPT_INJECTION]: The skill contains meta-instructions aimed at manipulating how sub-agents receive and interpret data. This includes techniques to ensure prompts are not fragmented by terminal newlines and instructions to write long prompts to /tmp/task.md to ensure they are followed as a single unit.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 18, 2026, 04:27 PM
Security Audit — agent-trust-hub — delegating-to-agents