v2.0.0 is a mechanism overhaul. v1.0.0 filtered AI slop out; v2.0.0 goes further and requires craft and liveliness in. The insight behind this version: slop and good design use the SAME techniques (gradients, glass, badges, motion). They differ only in purpose. A filter subtracts; liveliness has to be added.
What's new
- Part 3: Liveliness Toolkit (new). Three dials, ENERGY / RHYTHM / MOTION, each with 3 levels (not 1-10), so a model and a reviewer can actually check them. Positive levers: one focal point per screen, hierarchical contrast, whitespace as structure, one deliberate accent, an identity motif.
- Purpose-Gate tier. 12 rules moved from "forbidden" to "allowed if it serves hierarchy and the reason is written down." Techniques are not banned; technique without purpose is.
- Hard Gate stays absolute. 17 rules for honesty, function, and accessibility remain non-negotiable (fake stats, dead links, non-functional elements, etc.). Part 1 is now a diagnostic scan with an explicit carve-out for these.
- Quality Locks. 9 consistency rules. Total stays 38 rules (R-01 to R-38), just regrouped into three tiers.
- Design Read. The agent declares its direction before generating ("reading this as X for Y, in Z style, dial ..."). Ambiguous? It asks exactly one question. No direction? It labels the output a draft and defaults the dials to 1/1/1.
- 4-block Delivery Gate. Mandatory PASS/FAIL with evidence per item: Hard / Purpose-Gate / Liveliness / Craftsmanship & Quality Locks.
- 3-file system with DESIGN.md. ANTISLOP.md is the filter; DESIGN.md (yours) is the direction that makes a result feel alive; AGENTS.md routes the agent to both. A sterile result means direction was missing, not that the filter failed.
v1.0.0 stopped the slop. v2.0.0 is about what comes after the slop is gone: design that is honest, functional, consistent, and alive.