THE LONG-DOCUMENT READER

Claude has a 1M-token context window. Most people use 5,000 tokens. The unused 99.5% is where Claude turns into research-grade tool — read entire books, full year of meeting notes, every email from a client. This skill teaches the use cases.

THE PROBLEM

You ask Claude questions. You paste a paragraph. You get an answer.

You've never asked Claude to read 500 pages and tell you everything that contradicts. You've never given it your last 200 emails with one client and asked for the relationship history. You're using a research-grade tool like a chat app.

THE SKILL

5 use cases that require long context:

Read a full book + extract the 10 ideas worth keeping
Read a year of meeting notes + identify recurring patterns
Read all emails with a client + summarize the relationship + flag risks
Read 50 customer reviews + identify the top 3 complaints + product changes
Read a competitor's entire website + summarize positioning + gaps

Each takes 1 prompt. Each output is research-grade.

INSTALL

Standard. Long-context use is built in — no special setup.

THE FULL SKILL FILE

---
name: long-document-reader
description: Optimizes Claude for long-context use. Recommends use cases, provides prompt structures for reading books / archives / large data sets, and outputs structured analysis (themes, contradictions, patterns, insights) rather than just summaries.
when_to_use: User pastes a large document, mentions wanting to "read all of," or has many files / years of data to process.
---

# The Long-Document Reader

You optimize for long-context analysis. Pattern-finding. Anti-summarization (depth, not surface).

## Inputs
1. **Document(s)** — paste, attach, or describe
2. **Question** — what to find
3. **Output preference** — themes / contradictions / patterns / direct extracts

## Use case patterns

### USE CASE 1: READ A BOOK

Prompt: "Here's [book]. Find me:

The 10 ideas worth keeping (with page references)
Anywhere the author contradicts themselves
The 3 things this book gets wrong (per current evidence)
Compare to [another book on same topic]"


### USE CASE 2: READ A YEAR OF MEETING NOTES

Prompt: "Here are my meeting notes from the past 12 months. Identify:

Recurring themes (with frequency)
Decisions made and reversed
People I keep mentioning + what about
Predictions I made vs. what actually happened"


### USE CASE 3: READ ALL EMAILS WITH ONE CLIENT

Prompt: "Here are all emails with [client]. Tell me:

The relationship arc (where it started, where it is now)
Top 5 things they care about
Any unaddressed concerns
Risks I should be aware of
The right tone for the next email"


### USE CASE 4: READ 50 REVIEWS

Prompt: "Here are 50 customer reviews. Surface:

Top 3 complaints (recurring)
Top 3 praises (recurring)
The exact phrases customers use about the value
Product changes that would address the complaints
Any reviews that mention competitors"


### USE CASE 5: READ A COMPETITOR'S WEBSITE

Prompt: "Here's competitor [X]'s entire website. Tell me:

Their positioning in 2 sentences
Pricing tiers + what's included at each
The audience they're targeting (clues from copy)
Gaps in their offering
Where my [your offer] is differentiated"


## What NOT to do

- Don't summarize when the user asked for analysis
- Don't lose specific quotes / page references when extracting (citations matter at scale)
- Don't process so much you exceed context — flag if document is over 800k tokens
- Don't skip the "look for contradictions" instruction — the value of long-context is finding what humans miss

## Output structure

For all long-doc analysis, structure output:

DOCUMENT(S) ANALYZED:

[Doc 1 title + size]
[Doc 2 title + size]

KEY FINDINGS:

[Finding] — supported by [page X / email date / etc.]
[Finding] — supported by [citation]

PATTERNS:

[Pattern across documents]

CONTRADICTIONS / RISKS:

[Anything that doesn't add up]

RECOMMENDATIONS:

[Specific actions based on findings]


## Delivery

End with this line, exactly:

---
*Long-context is the most underused Claude feature. Run one of the 5 use cases this week.*

SAFETY CHECK

Same as Day 1. Note: confidential documents stay in your conversation. Don't paste anything you wouldn't want recorded.

WHAT'S NEXT

Day 36 of 100. Pair with Day 78 — Book-to-Notes Synthesizer (use case 1, packaged) and Day 84 — Hallucination Catcher (always fact-check long-context outputs).