Writes ralph loops for you that you can copy and paste

Generate ready-to-run shell loop commands for Claude Code, Gemini CLI, or Grok CLI, customized by AI tool, model, shell, complexity, and loop features.

This skill appears to genuinely generate loop commands, but you should be careful before using the generated commands. Things to consider: - Review any generated command before running it. The SKILL.md encourages use of flags that bypass permission prompts (e.g., --dangerously-skip-permissions, --yolo, 'auto-approves permissions'). Those remove interactive safety checks and can let a CLI perform actions without confirmation — avoid them unless you understand the consequences. - Do not include secrets in PROMPT.md or STATUS.md. The skill reads and pipes PROMPT.md into CLIs; anything in that file could be sent to the chosen AI service. - The instructions reference GROK_MODEL (and similar tool-specific configuration) but the skill metadata lists no required environment variables. If you rely on a particular model, set the env var yourself and verify its value. Treat this as a documentation gap. - Generated loops can be long-running or infinite. Test commands in a controlled/sandboxed environment, include stop-file or iteration limits, and avoid running in production systems without rate/timeout protection. - Ensure the underlying CLI tools (claude-code, gemini, grok) are installed from trusted sources before piping content to them. Given these inconsistencies and the encouragement to disable safety prompts, proceed cautiously — the skill is not obviously malicious, but its instructions increase risk if used carelessly.

Type: OpenClaw Skill Name: ralph-loop-writer Version: 1.0.0 The skill bundle is classified as benign. Its purpose is to generate automation loop commands for AI CLI tools (Claude, Gemini, Grok) and write them to a markdown file. The `SKILL.md` instructions are clear, transparent, and do not contain any prompt injection attempts against the agent. The allowed tools (`AskUserQuestion`, `Write`, `Read`) are used appropriately: `AskUserQuestion` for user interaction, and `Write` to create the output command file (`ralphcommand-YYYY-MM-DD-HHMMSS.md`). While the generated commands for the user may include flags like `--dangerously-skip-permissions` or `--yolo` for the target AI CLIs, these are features of those external tools for automation, and the skill itself does not execute them or instruct the OpenClaw agent to perform any high-risk actions, data exfiltration, or persistence mechanisms.