rgr

Turns a brief, discussion, or PRD into a Locked Intent Boundary - a small human-signed artifact that fixes intent before planning, so an instruction-following agent cannot silently redefine the goal or grade itself against its own rewrite. ONE mechanism for every task - destination deltas, invariants, non-substitutions, authorized change rights, proofs, stop/relock - with rigor scaled to blast radius, not a destructive/additive split. Use when turning research + discussion into a master PRD, before phasing/Linear issues, or for any build/remove/replace/migrate/refactor task. Triggers - "intent contract", "/intent-contract", "intent boundary", "lock the boundary", "write the master PRD", "before we plan this", "is this remove or replace". Derives the phased plan and runs a change-surface audit where every diff op (add/modify/delete/rename) must map to a signed authorization.

Use RGR for strict Red-Green-Refactor work in Claude Code or Codex when a code change should prove the test failed before implementation, protect that Red test from silent edits, prove Green with the same command, and finish with replay verification.