torsorlab / torsor-writing / write-topic-guide
8 of 13

write-topic-guide

a skill to take several papers or books on one topic, and write a single guide explaining it.

new to this? open a folder holding the sources you want synthesized in Claude Code or Codex, and say:

use the repository at https://github.com/torsor/claude-plugins and write a topic guide on from the sources in this folder

a Claude Code skill. give it several sources on one topic — papers, books, theses, including scanned ones with no text layer — and it reads them and writes a single guide explaining the idea across all of them.

where write-paper-guide companions one paper organized by that paper’s sections, this organizes by a conceptual spine you decide. it is for when no single source carries the exposition and something has to.

four things are particular to it. sources are pre-summarized — each one distilled by a dedicated subagent, in its own context, into a structured digest under source-notes/, and the guide is written from those notes rather than from the raw PDFs. scanned and image-only PDFs are handled by locating the relevant chapters and vision-reading them. you decide a concept spine up front, map every source onto it, and get back an honest list of what no source covers. and it is oriented to learning: calibrated to one named reader, developing missing prerequisites from scratch, and leaving the source-notes behind as a durable study artifact.

the framing rules carry over from the paper guide. third person about the sources — the authors prove, section 3 establishes. and a guide, never a contribution and never a survey of the field.

the shape

phase A   pre-summarize the sources     ->  source-notes/*.md  (kept)
phase B   reader + concept spine        ->  reader-profile.md, the spine, the gaps list
phase C   author and build              ->  the guide, four formats

invoke

/write-topic-guide how to think about descent, from these three sources: ...

produces

a guide directory in the house format — LaTeX source with PDF, HTML, EPUB and Markdown — with source-notes/ kept alongside as a deliverable in its own right.