What problem does it solve? New and existing projects often lack a well-structured global rules file, leaving AI coding agents without an always-on map of the codebase, ground rules, and working principles, or with a bloated generic /init output the agent ignores. ## Core Features & Use Cases - Greenfield derivation: Reads a PRD and/or architecture spec passed as arguments and derives rules from the settled technical decisions, treating the PRD only as product context. - Brownfield derivation: Derives rules from a primed codebase (via /prime-codebase) or a passed codebase-analysis document, requiring every rule to be traceable to a real file. - Lean composition workflow: Sorts content into always-on rules, on-demand .claude/references/ stubs, task plans, or deletions, and elicits working principles from the user rather than inventing them. - Use Case: After settling your stack and conventions with an architecture doc, run this skill with the spec path to produce a CLAUDE.md containing an architecture map, ground rules, commands, and agent-steering principles. ## Quick Start Ask the AI to run the rules-create-global skill with the path to your architecture spec to generate the project's root CLAUDE.md.