llm-html

CMDOP Skill — installieren und verwenden via CMDOP-Agent:

Bash

cmdop-skill install llm-html

Loading code...

LLM-optimiertes HTML-Reinigen: Hydratationsauszug, Tokenbudgets, mehrere Ausgabeformate.

Install

Bash

pip install llm-html

Loading code...

Schnellstart

Python

from llm_html import HTMLCleaner, CleanerConfig, OutputFormat

# Grundlegende Reinigung
  cleaner = HTMLCleaner()
  result = cleaner.clean(html)
  print(f"Reduzierung: {result.stats.reduktion_prozent}%")

# Hydratations-zuerst (SSR-Daten aus Next.js, Nuxt usw. extrahieren)
  if result.hydration_data:
    data = result.hydration_data
  else:
    cleaned = result.html

Loading code...

Convenience-Funktionen

Python

from llm_html import clean, clean_to_json, clean_html, clean_for_llm

# Schnelle Reinigung
  result = clean(html)

# JSON-Datei, wenn SSR-Daten verfügbar, ansonsten gereinigtes HTML
  data = clean_to_json(html)

# Pipeline mit voller Kontrolle
  result = clean_html(html, max_tokens=5000)
  result = clean_for_llm(html, output_format="markdown")

Loading code...

Ausgabeformate

Python

from llm_html import to_markdown, to_aom_yaml, to_xtree

md = to_markdown(html)
  aom = to_aom_yaml(html)
  xtree = to_xtree(html)

Loading code...

Downsampling

Token-budget-Zielsetzung mit D2Snap-Algorithmus:

Python

from llm_html import downsample_html, estimate_tokens

  tokens = estimate_tokens(html)
  if tokens > 10000:
    html = downsample_html(html, target_tokens=8000)

Loading code...

Semantische Chunking

Große Seiten in LLM-große Chunk aufteilen:

Python

from llm_html import SemanticChunker, ChunkConfig

  config = ChunkConfig(max_tokens=8000, max_items=20)
  chunker = SemanticChunker(config)
  result = chunker.chunk(soup)
  for chunk in result.chunks:
    process(chunk.html)

Loading code...

Shadow DOM

Flachere Web-Components für LLM-Sichtbarkeit:

Python

from llm_html import flatten_shadow_dom

  flat = flatten_shadow_dom(html)

Loading code...

Hilfsfunktionen

Python

from llm_html import html_to_text, extract_links, extract_images, json_to_toon

  text = html_to_text(html)
  links = extract_links(html, base_url="https://example.com")
  images = extract_images(html)
  toon = json_to_toon({"key": "value"})

Loading code...

Lizenz

MIT

Bash

cmdop-skill install llm-html

Bash

pip install llm-html

Python

from llm_html import HTMLCleaner, CleanerConfig, OutputFormat

# Grundlegende Reinigung
  cleaner = HTMLCleaner()
  result = cleaner.clean(html)
  print(f"Reduzierung: {result.stats.reduktion_prozent}%")

# Hydratations-zuerst (SSR-Daten aus Next.js, Nuxt usw. extrahieren)
  if result.hydration_data:
    data = result.hydration_data
  else:
    cleaned = result.html

Python

from llm_html import clean, clean_to_json, clean_html, clean_for_llm

# Schnelle Reinigung
  result = clean(html)

# JSON-Datei, wenn SSR-Daten verfügbar, ansonsten gereinigtes HTML
  data = clean_to_json(html)

# Pipeline mit voller Kontrolle
  result = clean_html(html, max_tokens=5000)
  result = clean_for_llm(html, output_format="markdown")

Python

from llm_html import to_markdown, to_aom_yaml, to_xtree

md = to_markdown(html)
  aom = to_aom_yaml(html)
  xtree = to_xtree(html)

Python

from llm_html import downsample_html, estimate_tokens

  tokens = estimate_tokens(html)
  if tokens > 10000:
    html = downsample_html(html, target_tokens=8000)

Python

from llm_html import SemanticChunker, ChunkConfig

  config = ChunkConfig(max_tokens=8000, max_items=20)
  chunker = SemanticChunker(config)
  result = chunker.chunk(soup)
  for chunk in result.chunks:
    process(chunk.html)

Python

from llm_html import flatten_shadow_dom

  flat = flatten_shadow_dom(html)

Python

from llm_html import html_to_text, extract_links, extract_images, json_to_toon

  text = html_to_text(html)
  links = extract_links(html, base_url="https://example.com")
  images = extract_images(html)
  toon = json_to_toon({"key": "value"})

llm-html

llm-html

Install

Schnellstart

Convenience-Funktionen

Ausgabeformate

Downsampling

Semantische Chunking

Shadow DOM

Hilfsfunktionen

Lizenz

Yorumlar