deep-reader

Name: deep-reader
Author: tronghieu

37stars20forks

How to install deep-reader?

tronghieu/agent-skills/deep-reader

$npx -y skills add tronghieu/agent-skills --skill deep-reader

Installs into the current project.

Summary

Deep-read long books and papers using Mortimer Adler's inspectional / analytical / syntopical method, the Recite step from SQ3R, and S. Keshav's three-pass method for papers, with page-anchored notes as external memory so the agent never has to load the whole text into one contex

Files · 1

View on GitHub

SKILL.md

1	# Deep Reader
2
3	Read long books and papers the way a careful human researcher does: skim for
4	structure first, read deeply chapter by chapter with notes as you go, then
5	synthesize from the notes — never the raw text — at the end. The method comes
6	from Adler's How to Read a Book, SQ3R's Recite step, and Keshav's three-pass
7	method for papers.
8
9	## Why multi-pass, why notes
10
11	An LLM's attention dilutes over long contexts, and it loses track of material
12	buried in the middle of a big prompt ("lost in the middle"). Loading a
13	500-page book into one context and asking questions about it produces answers
14	that quietly forget chapter 3 by the time you're discussing chapter 12.
15
16	The fix is the same one human researchers use: never hold the whole book in
17	working memory at once. Read it in passes — first for structure, then chapter
18	by chapter for content — and externalize what you learn into note files as you
19	go. The notes become the durable memory; the book itself only needs to be
20	back in view when you're actively reading a specific chapter or verifying a
21	specific claim. Page numbers are the coordinate system that makes every claim
22	traceable back to the source, in this session or a later one.
23
24	## Two modes
25
26	- overview — Adler's inspectional pass only: a book map plus a
27	goal-directed summary of the chapters that matter most to the reader's
28	purpose. Fast, no per-chapter notes.
29	- study — the full pipeline: reading purpose → inspectional map →
30	chapter-by-chapter analytical notes with Recite verification → hierarchical
31	synthesis answering Adler's four questions → mechanical quote verification.
32
33	The user picks (or you propose one from `mode_hint` + their stated purpose —
34	see Step 1).
35
36	## Workspace layout
37
38	Everything the pipeline produces lives in a workspace directory, by default a
39	sibling of the source file, so it survives across sessions:
40
41	```
42	<slug>-notes/
43	├── source.txt # full text with [[page N]] markers — the coordinate system
44	├── chapters.tsv # confirmed structure: chNN<TAB>from<TAB>to<TAB>title
45	├── chapters/ # per-chapter files cut by split-chapters.sh
46	│ └── ch01-<slug>.txt
47	├── map.md # inspectional pass output
48	├── terms.md # key-terms ledger, cross-chapter
49	├── notes/
50	│ └── ch01-<slug>.md # one analytical note per chapter (papers: sec01-<slug>.md)
51	└── synthesis.md # final synthesis + verification log
52	```
53
54	This workspace, not the SKILL.md scripts, is what makes a later session able
55	to answer a follow-up question without re-reading the book (see "Answering
56	from existing notes" below).
57
58	## Scripts at a glance
59
60	All seven scripts live in `scripts/` under this skill. Run them either as
61	`bash ./scripts/<name>.sh ...` from the skill directory, or as
62	`bash /mnt/skills/user/deep-reader/scripts/<name>.sh ...` on claude.ai — both
63	forms invoke the same script, use whichever matches where you're running.
64
65	\| Script \| Arguments \| Purpose \|
66	\| --- \| --- \| --- \|
67	\| `prepare-text.sh` \| `<input-file> [workspace-dir]` \| Convert source to paged `source.txt`, create the workspace \|
68	\| `extract-structure.sh` \| `<source.txt>` \| Heuristic outline seed (a starting guess, not ground truth) \|
69	\| `read-pages.sh` \| `<source.txt> <from> [to]` \| Print a page or page range, with `[[page N]]` markers intact \|
70	\| `search-book.sh` \| `<source.txt> <pattern> [context-lines]` \| Grep the whole book, page-annotated results \|
71	\| `split-chapters.sh` \| `<source.txt> <chapters.tsv>` \| Cut confirmed chapters into standalone files \|
72	\| `build-diagram.sh` \| `<workspace-dir> [--append]` \| Mechanical Mermaid mindmap of the structure \|
73	\| `verify-quotes.sh` \| `<workspace-dir>` \| Catch fabricated quotes and wrong page citations \|
74
75	## Step 0 — Prepare
76
77	Locate the source file, then convert and page it:
78
79	```bash
80	bash ./scripts/prepare-text.sh path/to/book.pdf
81	# or on claude.ai:
82	bas

People who liked this also viewed

6 picks

Type

deep-reader

37stars20forks

View on GitHub

How to install deep-reader?

tronghieu/agent-skills/deep-reader

$npx -y skills add tronghieu/agent-skills --skill deep-reader

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Deep Reader
2
3	Read long books and papers the way a careful human researcher does: skim for
4	structure first, read deeply chapter by chapter with notes as you go, then
5	synthesize from the notes — never the raw text — at the end. The method comes
6	from Adler's How to Read a Book, SQ3R's Recite step, and Keshav's three-pass
7	method for papers.
8
9	## Why multi-pass, why notes
10
11	An LLM's attention dilutes over long contexts, and it loses track of material
12	buried in the middle of a big prompt ("lost in the middle"). Loading a
13	500-page book into one context and asking questions about it produces answers
14	that quietly forget chapter 3 by the time you're discussing chapter 12.
15
16	The fix is the same one human researchers use: never hold the whole book in
17	working memory at once. Read it in passes — first for structure, then chapter
18	by chapter for content — and externalize what you learn into note files as you
19	go. The notes become the durable memory; the book itself only needs to be
20	back in view when you're actively reading a specific chapter or verifying a
21	specific claim. Page numbers are the coordinate system that makes every claim
22	traceable back to the source, in this session or a later one.
23
24	## Two modes
25
26	- overview — Adler's inspectional pass only: a book map plus a
27	goal-directed summary of the chapters that matter most to the reader's
28	purpose. Fast, no per-chapter notes.
29	- study — the full pipeline: reading purpose → inspectional map →
30	chapter-by-chapter analytical notes with Recite verification → hierarchical
31	synthesis answering Adler's four questions → mechanical quote verification.
32
33	The user picks (or you propose one from `mode_hint` + their stated purpose —
34	see Step 1).
35
36	## Workspace layout
37
38	Everything the pipeline produces lives in a workspace directory, by default a
39	sibling of the source file, so it survives across sessions:
40
41	```
42	<slug>-notes/
43	├── source.txt # full text with [[page N]] markers — the coordinate system
44	├── chapters.tsv # confirmed structure: chNN<TAB>from<TAB>to<TAB>title
45	├── chapters/ # per-chapter files cut by split-chapters.sh
46	│ └── ch01-<slug>.txt
47	├── map.md # inspectional pass output
48	├── terms.md # key-terms ledger, cross-chapter
49	├── notes/
50	│ └── ch01-<slug>.md # one analytical note per chapter (papers: sec01-<slug>.md)
51	└── synthesis.md # final synthesis + verification log
52	```
53
54	This workspace, not the SKILL.md scripts, is what makes a later session able
55	to answer a follow-up question without re-reading the book (see "Answering
56	from existing notes" below).
57
58	## Scripts at a glance
59
60	All seven scripts live in `scripts/` under this skill. Run them either as
61	`bash ./scripts/<name>.sh ...` from the skill directory, or as
62	`bash /mnt/skills/user/deep-reader/scripts/<name>.sh ...` on claude.ai — both
63	forms invoke the same script, use whichever matches where you're running.
64
65	\| Script \| Arguments \| Purpose \|
66	\| --- \| --- \| --- \|
67	\| `prepare-text.sh` \| `<input-file> [workspace-dir]` \| Convert source to paged `source.txt`, create the workspace \|
68	\| `extract-structure.sh` \| `<source.txt>` \| Heuristic outline seed (a starting guess, not ground truth) \|
69	\| `read-pages.sh` \| `<source.txt> <from> [to]` \| Print a page or page range, with `[[page N]]` markers intact \|
70	\| `search-book.sh` \| `<source.txt> <pattern> [context-lines]` \| Grep the whole book, page-annotated results \|
71	\| `split-chapters.sh` \| `<source.txt> <chapters.tsv>` \| Cut confirmed chapters into standalone files \|
72	\| `build-diagram.sh` \| `<workspace-dir> [--append]` \| Mechanical Mermaid mindmap of the structure \|
73	\| `verify-quotes.sh` \| `<workspace-dir>` \| Catch fabricated quotes and wrong page citations \|
74
75	## Step 0 — Prepare
76
77	Locate the source file, then convert and page it:
78
79	```bash
80	bash ./scripts/prepare-text.sh path/to/book.pdf
81	# or on claude.ai:
82	bas

People who liked this also viewed

6 picks

Type