What problem does it solve? Signed product requirement documents accumulate verbose prose that duplicates what the built code already expresses, making the PRD expensive to maintain and prone to drift. This Skill runs the once-per-tier compression sweep that reduces a tier's fat prose down to intent, acceptance criteria, and pointers while preserving the original text verbatim in an archive. ## Core Features & Use Cases - Gated compression sweep: Runs only after the tier's human taste review (HR-item) closes, and drafts surgical subsection-level edits across PRD sections rather than whole-section rewrites. - Verbatim archival: Moves demoted text unchanged into a dated archive file with a forward pointer, so the reasoning behind decisions is never lost. - Human-approved landing: Files the sweep as a reel-back plan plus an HR-item, and lands the result as a single commit only after human approval. - Use Case: After the T0 milestone demo's taste review closes, invoke this Skill to compress PRD sections 2-6 into intent plus acceptance criteria, archive the original prose, and insert generated regions where the build matches end-state intent. ## Quick Start Ask the AI to run the prd-compress sweep for tier T0 now that its taste review HR-item has closed.
