parity extract
Single-site, AI-ready component extraction — HTML, computed styles, screenshots, assets per component.
parity extract is the lower-level, single-site capture: it snapshots one site's UI components (header, footer, nav, shelves, hero, minicart, …) into a structured, AI-ready bundle — no prod × cand comparison.
parity extract --url https://loja.com \
--pages /,category-auto,pdp-auto --format bothUse it when there's no source code to read and you want the raw component inventory to feed an AI agent. (parity migrate builds on the same primitives and adds theme, target playbooks, VTEX content, and a visual report.)
What it captures
A heuristic pass always runs (no LLM required): semantic tags (header/footer/nav), Deco [data-section], class heuristics (shelf/carousel/minicart), and a geometry pass for hero/banner. An optional LLM pass relabels the detected roles.
Per component: rendered HTML, computed styles, a cropped screenshot, and asset/link/text inventories.
Flags
| Flag | Default | What it does |
|---|---|---|
--url | required | Site to extract from |
--pages | / | Literal paths and/or category-auto / pdp-auto |
--components | all | Role allowlist (e.g. header,footer,shelf) |
--viewport | mobile | mobile | desktop | tablet |
--format | both | md | json | both |
--out | ./parity-extract | Output directory |
--no-llm | LLM on | Skip the relabel pass |
Output
parity-extract/<host>/<timestamp>/ — index.md (overview), manifest.json (full bundle), and components/<role>/ (README + component.html + styles.json + screenshot.png).
Keep parity-extract/ gitignored — it contains a real store's content.
See also
parity migrate— the full migration bundle