Skip to main content
Design the look, informed by the competition — not generic taste. Positioning grounds the words; the design pillar grounds the look. From a finished demand report it authors a design system: an aesthetic direction, the one thing someone should remember, and one SAFE/RISK choice per design dimension (typography, color, layout, spacing, motion, decoration), plus a DESIGN.md source of truth and a preview page. Unlike the rest of metalworks, design is taste, so grounding here is directional, not cited: the competitive landscape informs the bet (“rivals skew serif → lean serif, or break to sans”); it does not cite it. Two honesty signals carry the weight instead — every choice is labelled SAFE (category baseline) or RISK (a deliberate departure), and the system records its grounding tier.

The grounding tier (read this first)

The system records how grounded the look actually is, so a confident design that never saw a competitor can’t masquerade as a grounded one:

Taste presets

A small, curated set of opinionated directors — pick one with taste= (Python/MCP) or --taste (CLI). The chosen preset is recorded on system.taste and drives the preview/logo-picker chrome. The same report under two presets yields a visibly different system.

What you get back

The choices carry no evidence_refs by design — design grounding is directional, not cited. This pillar authors a system, not pixels: hand the DESIGN.md to your build step to apply it.

A real teardown needs the browser

The renderer tier — a real screenshot-and-style teardown of competitor sites — needs the browser extra and Chromium:
Without it the pillar still works, just at the web or model_knowledge tier — always clearly labelled. On a server, set FIRECRAWL_API_KEY to render without a local browser.