doc-splitter_skill

Overview

This skill splits very large documentation sites into focused sub-skills and optionally creates an intelligent router to direct queries. It is designed for massive doc collections (10k+ pages) where single-skill scale or navigation becomes impractical. The goal is predictable, auditable splits with safety checks and retryable recovery steps.

How this skill works

It first estimates total page count and analyzes natural category boundaries from the documentation structure. You choose a split strategy (category, size, router, or hybrid) and the skill generates sub-skill configurations, scrapes sub-sites in parallel, and produces an optional router that maps queries to the right sub-skill. Checkpoints and recovery protocols capture progress and enable targeted retries on failures.

When to use it

• Documentation exceeds ~10,000 pages and single-skill performance degrades
• Site has clear topical sections that can become independent skills
• You need deterministic routing for user queries across many topics
• Preparing documentation for multi-agent or microservice workflows
• When you require repeatable split plans and checkpointed progress

Best practices

• Run a full size estimation before any split and confirm category detection
• Prefer router + category strategy for 10k–30k+ pages to preserve UX
• Ask the user when category boundaries or target sizes are unclear — never guess
• Use URL patterns and a catch-all category to avoid orphan pages
• Validate sub-skill coverage and run a small user routing test before production

Example use cases

• Splitting a 32k-page engine manual into scripting, 2D, 3D, physics, and networking sub-skills
• Creating a router for a cloud provider docs site so queries route to compute, storage, or networking agents
• Uniformly partitioning an API reference into size-limited sub-skills for parallel scraping
• Recovering and re-splitting a problematic category after detection of overlapping URL patterns
• Generating sub-skill configs for staged scraping and CI-driven validation