ralph-technique

npx skills add majiayu000/claude-skill-registry --skill ralph-technique --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

# Ralph Wiggum Technique The Ralph technique is a minimal prompting approach that enables autonomous, loop-based agent execution. Named after the Simpsons character, it embraces simplicity and iterative refinement. ## Core Philosophy > "Ralph is deterministically bad in an undeterministic world." **Key insight:** Simpler prompts (~40-50 lines) often outperform verbose prompts (~200+ lines). Overly detailed prompts can make agents "slower and dumber." ## The Ralph Loop In its purest form, Ralph is a bash loop: ```bash while :; do cat PROMPT.md | claude-code ; done ``` The agent runs continuously, making incremental progress. Failures are expected and corrected through iteration. ## Signs on the Playground When Ralph makes mistakes, don't blame the tools—add "signs": ``` Ralph builds playground → Falls off slide → Add sign: "SLIDE DOWN, DON'T JUMP" → Ralph sees sign next time → Behavior improves ``` **Translation:** When an agent fails, add a concise constraint to the prompt. Don't explain why—just state the rule. ## Minimal Prompt Pattern ```markdown # {Agent} - {Role} You are {Agent}. Your job is to {primary task} in `task/`. ## Constraints - {Essential constraint 1} - {Essential c

while :; do cat PROMPT.md | claude-code ; done