Claude Config

Use when starting any non-trivial feature, design, or behavior change - before writing code or creating tasks. Guides collaborative design exploration through focused questioning to produce a validated spec.

Use when setting up autonomous agent execution loops for a project. Takes a completed design doc and produces IMPLEMENTATION.md, PROMPT files per loop, progress trackers, cross-validation, and ralph.sh runner. Use after brainstorming/design is done, before coding starts.

Use after completing non-trivial work — new features, refactors, multi-file changes, non-trivial bug fixes — before reporting the task complete. Guides parallel review across six lenses (performance & memory, code quality & maintainability, architecture, testing, security, codebase consistency) by spawning focused review agents. Validated findings are fixed directly; architectural red flags and changes to product behavior surface for user discussion. Skip for typos, one-line fixes, pure renames, or config-value tweaks.