ralphmode

Warn

Audited by Gen Agent Trust Hub on Mar 20, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides instructions for using flags that bypass security sandboxes and approval prompts.
  • Evidence: Recommendations for --dangerously-skip-permissions (Claude), --dangerously-bypass-approvals-and-sandbox (Codex), and --yolo (Gemini) which disable core safety features.
  • [COMMAND_EXECUTION]: The skill instructs users to modify global configuration files and create persistent hooks.
  • Evidence: Directions to modify ~/.claude/settings.json and ~/.gemini/settings.json and create scripts in ~/.claude/hooks/ and ~/.gemini/hooks/.
  • [REMOTE_CODE_EXECUTION]: The skill provides full shell script templates and directs the user to make them executable.
  • Evidence: Templates for ralph-safety-check.sh and ralph-tier1-check.sh that are integrated into the agent's execution flow via configuration hooks.
  • [PROMPT_INJECTION]: The skill implements "prompt contracts" as a primary safety measure for certain platforms.
  • Evidence: Instructions for Codex and OpenCode that rely on the agent following natural language rules to ask for permission (CHECKPOINT_NEEDED) rather than technical enforcement.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 20, 2026, 07:00 AM