Agent skill · writing content · erichowens
very-long-text-summarization
Summarizes very long texts (books, handbooks, biographies, codebases) using hierarchical multi-pass extraction with cheap model armies. Produces structured knowledge maps, not just summaries.
Why this skill is useful
Adds a structured multi-pass extraction method for summarizing large texts that the AI wouldn't generate on its own.
What it needs
About 5k tokens when loaded. Last updated 2026-07-14. 180 stars on the source repository.
What this skill does
Very Long Text Summarization Processes texts too large for a single context window using hierarchical multi-pass extraction with armies of cheap models. Produces structured knowledge maps, indexed summaries, and skill drafts — not just prose compression. --- When to Use ✅ Use for: Professional handbooks and textbooks (100-1000+ pages) Career biographies and memoirs (extracting expertise patterns) Large codebases (architecture-level understanding) Research paper collections (synthesizing findings across papers) Any text exceeding a single context window (~100K tokens) ❌ NOT for: Short documents (<10 pages) — just read them directly Real-time conversation summarization (use auto-compact patterns) Code documentation generation (use technical-writer) Simple TL;DR requests (not worth the multi-pass overhead) --- Architecture: Three-Pass Hierarchical Extraction Pass 1: Chunked Extraction (Haiku Army) Split the document into overlapping chunks (~4K tokens each, 500 token overlap). Deploy one Haiku call per chunk in parallel. Each extracts: Cost: ~$0.001 per chunk. A 300-page book (~150K tokens) = ~38 chunks = ~$0.04 total for Pass 1. Parallelism: All chunks run simultaneously. A 300-page book completes Pass 1 in ~3 seconds (wall clock), not 3 minutes. Pass 2: Synthesis (Sonnet) Feed all Pass 1 extractions into one or more Sonnet calls. Sonnet merges, deduplicates, and structures the knowledge. Cost: ~$0.02-0.05 depending on extraction volume. The index preserves traceability back to specific book sections. Pass 3: Refinement (Opus, Optional) For skill-draft output mode: Opus takes the knowledge map and produces a SKILL.md following the skill-architect template. This is the "crystallize skill from handbook" pipeline. Cost: ~$0.10. Only run when the output is a skill draft. --- Chunking Strategy Semantic Chunking (Preferred) Split on document structure — chapter boundaries, section headings, paragraph breaks. Preserves semantic coherence within each chunk. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills erichowens/very-long-text-summarization