Reimagine It Extract Agent Skill

/reimagine-it extract

Parent: ../SKILL.md — the full Content-Derived Design engine. This sub-skill handles only the extraction surface — the honesty layer the whole method stands on.

The redesign is only as honest as its extraction. This verb emits exactly what

the engine reads, as JSON, without generating anything: title, profile,

density, derived palette, anchors, proper nouns, headings, dates, numbers,

emails, links, and source hex colors. Every extracted fact exists in the

source — property-tested against hostile and malformed input.

Usage

npx reimagine-it extract -i page.html -o signals.json
cat page.html | npx reimagine-it extract -o -        # stdin → stdout
npx reimagine-it extract -i page.html --full -o signals.json  # + prose

Payload shape (default):

{
  "source": "page.html",
  "title": "…",
  "profile": "restaurant | saas | civic | …",
  "density": "sparse | medium | dense",
  "palette": { "ground": "…", "accent": "…", "muted": "…", "surface": "…", "ink": "…" },
  "anchors": ["…"],
  "properNouns": ["…"],
  "headings": ["…"],
  "dates": ["…"],
  "numbers": ["…"],
  "emails": ["…"],
  "links": ["…"],
  "sourceHex": ["#6B3A2A"],
  "counts": { "paragraphs": 4, "listItems": 16 }
}

--full adds paragraphs and items (the raw prose arrays).

Procedure

• Run it before any redesign when the user asks "what would you even use

from this page?" — the JSON answers with the engine's actual inputs.

• To audit honesty: every dates, numbers, emails, and links entry

must be findable in the source. Nothing else is allowed to appear.

• Use it for migration/research: the payload is a stable, typed summary of

a page's content — no redesign attached.

Must not

• Add facts not present in the payload to any downstream summary
• Present the derived palette as colors the source *declared* — hex colors

in sourceHex were in the source; the rest is derivation

• Round or "clean up" extracted numbers and dates; they are quotes

Part of reimagine-it — MIT.