doc-reader

The doc-reader skill provides a structured framework for consuming technical documentation without overwhelming your context window. Designed for developers and technical writers, it enables precise retrieval of information from APIs, SDKs, libraries, and third-party services. By prioritizing semantic search via the Model Context Protocol (MCP) and leveraging AI-friendly resources like llms.txt, the skill minimizes token waste while maximizing comprehension of complex technical requirements.

• Utilizes MCP search tools to perform semantic lookups across documentation sites for exact function, endpoint, or configuration details.

• Supports the identification and parsing of llms.txt and llms-full.txt files to ingest structured summaries or comprehensive site content.

• Implements smart fetching strategies including markdown URL variants for cleaner parsing and localized grep-based searches when semantic indexing is unavailable.

• Evaluates content length before ingestion, providing safety guardrails to prevent excessive context consumption or truncated information retrieval.

• Facilitates exploration of documentation sites by guiding the user to relevant HTML-based navigation when a high-level overview or interactive UI component is required.

• Invoke this skill whenever you need to integrate a new library, debug an unfamiliar error code, or verify implementation requirements against official documentation.

• Always check for existing MCP servers at documentation root endpoints to establish more efficient, context-aware communication with the service.

• When encountering long documentation, prefer targeted searches or partial loading over broad ingestion to maintain context efficiency.

• Inputs typically include documentation URLs or specific search queries; outputs provide technical summaries, code references, and direct paths to the required information.

• Constraints: Reliability depends on the availability of AI-native resources (llms.txt/MCP) on the host site; fallback to HTML scraping is supported but may be prone to noise or truncation for extremely large documents.