Ralph

This is specification-first development that refuses to let you write code until you've answered the hard questions. It runs Socratic interviews to expose hidden assumptions, then locks into autonomous loops (Interview → Seed → Execute → Evaluate → Evolve) until your acceptance criteria actually pass. The three-stage verification gates check mechanical correctness, semantic alignment, and optional multi-model consensus for drift detection. When you hit a wall after repeated attempts, it throws five lateral-thinking personas at the problem to break the stagnation. Works across Claude Code, Codex CLI, Gemini CLI, and OpenCode with unified syntax. The whole approach assumes your initial prompt is probably underspecified, which is usually true but feels like eating your vegetables before dessert.

Install to Claude Code

npx -y skills add supercent-io/skills-template --skill ralph --agent claude-code