Reading 1,400 Prompts as a Map: What the PCH Catalog Revealed

At first, this looked like a simple cataloging job. Convert the books to text, extract the prompts, put them into a spreadsheet, and move on. But after scanning 11 source TXT files and extracting 1,400 prompt candidates, the work started to feel less like collecting prompts and more like drawing a map.

A useful prompt is rarely just one sentence. Some prompts are starting commands. Some anchor the context. Some behave more like a harness that controls repeated execution. That is why I decided not to publish the source prompts as a prompt dump. The internal catalog preserves IDs, source books, and line ranges; the public notes read the patterns.

The first numbers are simple.

Item	Count
Converted source TXT files	11
Extracted prompt candidates	1,400
Prompt N blocks	1,001
Largest category	Creative Writing & Storytelling
Practical categories	Content, Marketing, Product, Research, Education, Coding

The distribution is the interesting part. Creative writing prompts dominate the set, while practical work prompts form smaller clusters. At first that looks uneven. Read differently, it becomes useful. Creative prompts are training grounds for world, character, event, emotion, and constraint. Practical prompts often start as one-line commands that need context and verification added later.

Reading Prompts in Three Layers

PCH stands for Prompt, Context, and Harness. Two prompts may look similar as text while serving different roles.

Layer	Question	Signal in the catalog
Prompt	What should the model do?	Write, summarize, analyze, generate, transform
Context	What must the model know?	Audience, data, situation, world, constraints
Harness	How will we repeat and verify it?	Output contract, pass/fail, stages, checklist

Many prompts begin at the Prompt layer: write a scene, generate blog ideas, explain code, summarize a document. But practical use quickly demands Context. Who is the audience? What material should the model rely on? What depth is expected? What should be avoided?

Reuse requires Harness. How do we judge the answer? Should the output be prose, a table, or JSON? What counts as failure? Without that layer, prompting depends too much on luck.

A Catalog Is an Operating Board

Collecting prompts feels useful until the collection becomes too large. Past 100 prompts, retrieval becomes work. Past 1,000, the main visible fact is simply that there are many prompts. The purpose of this catalog is not storage. It is operation.

The useful columns were not just the original prompt.

• Original ID
• Source book and line range
• Extraction pattern
• MECE category
• PCH layer
• Target type
• Intent summary
• PCH-upgraded prompt

With that structure, prompts stop being things to search and become things to design. The question shifts from “find me a marketing prompt” to “show me marketing prompts whose context is weak, then extend them into a harness.” That difference matters.

First Takeaway

Prompt engineering is not mainly about writing more impressive sentences. It is about making the work contract explicit. What the model should do, what context the user must supply, and how the result will be verified need to meet inside the same prompt.

I want to read this catalog in three directions next.

1. How creative prompts build worlds through constraints.
2. Why practical prompts need to grow from one-line commands into harnesses.
3. How a prompt catalog can connect to a personal LLM Wiki.

A prompt is not just a sentence. It is an interface. This catalog is the first map for redesigning that interface.