Sign InOpen Brain
GitHubGitHub RepoNeeds Review

virgiliojr94/book-to-skill

book-to-skill compiles books and document sets into on-demand agent skills, reducing repeated context loading while preserving chapter-level references and reusable decision rules.

GitHub
Open Source Open MarkdownOpen JSON
Source Summary

book-to-skill converts PDFs, ebooks, document folders, or globs into a shared SKILL.md, a core index, and on-demand chapter files. Its tests report **24–51× fewer tokens** than repeatedly placing a full book in context.

Practical Implication

Use this pattern for references you revisit: internal docs, standards, research clusters, or technical books. The generated structure front-loads mental models and loads only the relevant chapter, while the same skill format works in Claude Code, Amp, and GitHub Copilot CLI.

Agent-Ready Context
book-to-skill converts PDFs, ebooks, document folders, or globs into a shared SKILL.md, a core index, and on-demand chapter files. Its tests report **24–51× fewer tokens** than repeatedly placing a full book in context.

Use this pattern for references you revisit: internal docs, standards, research clusters, or technical books. The generated structure front-loads mental models and loads only the relevant chapter, while the same skill format works in Claude Code, Amp, and GitHub Copilot CLI.

The repository says conversion costs roughly **$1 per book**, and repeated use is where amortization helps. Chapter detection expects explicit numbered headings; one-off reading, broad library search, and poorly segmented sources may be better served by other approaches.
Connected Context · Feed7 Judgment

book-to-skill makes progressive disclosure concrete for stable, repeatedly consulted documents: preserve a shared mental model and index, then load chapters on demand. Its reported token reduction supports this narrow reuse case, but does not overturn evidence that full cached history can win for conversational recall. Conversion cost, heading requirements, and weak support for one-off or cross-library search define when richer retrieval or raw-history baselines remain preferable.

LLM Knowledge Bases: a practical guide — Ben Holmes, WarpBoth preserve source material while deriving lightweight indexes for navigation; the knowledge-base candidate adds taxonomy, backlinks, synchronization, and maintenance requirements for broader collections.Context Engineering in 2026 — Louis-François Bouchard, Omar Solano & Samridhi Vaid, Towards AIThe tutor result challenges token reduction as a universal goal, so chapter loading should be compared with full cached context on the actual workload.WTF Is the Context Layer? The Missing Infrastructure for Production Agents — Prukalpa SankarA generated shared skill is a concrete portable context artifact, reinforcing the candidate’s preference for versioned context shared across changing agent harnesses.RLM: Recursive Language Models for Large Codebases - Shashi, Superagentic AIBoth keep large sources outside the main prompt and return bounded relevant evidence; book-to-skill precomputes chapter structure, while RLM-style inspection queries external data dynamically.
Context Map
contextcodingresearch#skills#context-engineering#retrieval
Uncertainty
The repository says conversion costs roughly **$1 per book**, and repeated use is where amortization helps. Chapter detection expects explicit numbered headings; one-off reading, broad library search, and poorly segmented sources may be better served by other approaches.