context-eng_skill

Overview

This skill provides a Context Engineering Framework that autonomously extracts and structures unstructured information from websites, documents, and raw text into AI-interpretable formats. It standardizes agent behavior, input schemas, error handling, and output formats so teams can reliably convert heterogeneous sources into YAML/JSON or Markdown with front matter. The framework is designed for end-to-end automation: discovery, extraction, structuring, and persistence.

How this skill works

The skill defines a generic autonomous agent specification that orchestrates source fetching, hierarchical structure extraction, relevance grouping, and persistent output generation. It includes input schema definitions for source lists and processing options, pluggable tool roles (fetcher, extractor, URL discovery, filesystem manager), and an autonomous workflow loop that crawls, extracts, and adds newly discovered sources. Error handling and success criteria are built in so partial results are retained and performance targets are enforced.

When to use it

• Converting documentation websites into structured knowledge for LLM context
• Extracting and summarizing content from mixed inputs (URLs, files, raw text)
• Bootstrapping domain-specific knowledge bases or searchable indexes
• Automating discovery and collection of related content across domains
• Preparing curated, structured context for multi-agent workflows

Best practices

• Start with a manual source analysis template to map navigation, URL patterns, and metadata before automating crawling
• Define clear processing options: crawl depth, domain patterns, granularity, and output format up front
• Use hierarchical extraction settings (heading levels) to preserve logical structure for downstream models
• Limit crawl scope and set page caps to control cost and runtime
• Log partial outputs and enable graceful degradation so successful extractions are not lost on errors

Example use cases

• Structure a product documentation site into Markdown files with YAML front matter and an index for a docs search engine
• Ingest a corporate blog and extract topic-level summaries and tags to feed a retrieval-augmented generation (RAG) pipeline
• Crawl partner sites to discover related whitepapers, extract sectioned summaries, and store as JSON for downstream analytics
• Convert a folder of mixed files into a consistent knowledge tree for a team’s internal knowledge base