extract-html
Extract structured JSON from HTML using Schematron3B, deterministic table extraction, and Vision-based media OCR.
来源信息
- 仓库
- grahama1970/agent-stack-public
- 最近来源活动
- 2026年9月24日 15:51
- 检测到的 SKILL.md 语言
- 英语
- 星标
- 0
- 分支
- 0
安装方式
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
检查来源文件
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
文件资源管理器
19 个文件正在显示 SKILL.md
SKILL.md
来源说明 · 只读预览- name
- extract-html
- description
- Extract structured JSON from HTML using Schematron3B, deterministic table extraction, and Vision-based media OCR.
- triggers
- ["extract html","html to json","scrape html","table extraction"]
- provides
- ["extract-html"]
- composes
- ["task-monitor","agentic-evals"]
- disciplines
- ["extraction"]
# Extract-HTML Skill
A robust skill for converting HTML documents into strictly valid JSON based on a user-provided JSON Schema.
## Capabilities
1. **Schema Compliance**: Guarantees output conforms to the provided JSON Schema (using Schematron-3B + validation loop).
2. **Deterministic Tables**: Extracts HTML tables using `pandas.read_html` and injects them as context, preventing hallucination of data.
3. **Media Text Extraction**: Identifies images, filters by pixel size, and optionally uses a Vision API (OpenAI-compatible) to extract text/OCR.
4. **Self-Correction**: Validates model output and retries with error feedback if schema validation fails.
## Usage
### Basic Conversion (Local Only)
```bash
./run.sh convert \
--html input.html \
--schema target.schema.json \
--out result.json
```
### Advanced (With Vision & Remote Fetch)
```bash
./run.sh convert \
--html input.html \
--schema target.schema.json \
--out result.json \
--fetch-remote-media \
--vision-api-base "https://glhf.chat/api/openai/v1" \
--vision-api-key "sk-..." \
--vision-model "gpt-4o-mini"
```
## Options
- `--max-attempts <int>`: Number of self-correction retries.
- `--extract-tables / --no-extract-tables`: Toggle deterministic table extraction.
- `--extract-media-text`: Enable image processing.
- `--min-image-px`, `--max-image-px`: Filter images by size.
- `--include-sections`: detailed H1-H6 hierarchy in context.
## Dependencies
- Ollama running `schematron-3b` (or compatible model).
- Python 3.11+
- See `pyproject.toml` for python deps.
在 GitHub 查看