ralphmode
Fail
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: Provides detailed configuration templates and CLI commands specifically designed to disable built-in security mechanisms, including sandbox environments and mandatory user approval prompts, across multiple platforms (e.g., '--dangerously-skip-permissions' for Claude Code and 'sandbox_mode = "danger-full-access"' for Codex CLI).
- [PROMPT_INJECTION]: Contains explicit instructions for the AI agent to adopt custom 'checkpoint' protocols (e.g., 'CHECKPOINT RULE') that are intended to override or replace the standard safety decision-making processes provided by the platform.
- [EXTERNAL_DOWNLOADS]: References and links to official documentation and configuration guides from well-known technology providers such as Anthropic, OpenAI, and Google.
Recommendations
- AI detected serious security threats
Audit Metadata