CZON

Basic Rules for AI Analysis Report Generation

Prompt Engineering

👤 AI developers, document analysts, users of automated report generation tools, and personnel who need to process large numbers of Markdown files and generate structured reports.
This document outlines the basic rules for AI analysis report generation, with core content covering the document reading process (phased file list acquisition, batch reading and summarization, context management, completeness verification, and report generation), general rules (such as the fact-based principle, citation link specifications, header format, and time weighting), and prohibited behaviors. It emphasizes strict adherence to the process, reading all files one by one, generating reports based on facts, and provides specific operational guidelines and examples to ensure the systematicity, accuracy, and completeness of reports.
  • ✨ Must read all Markdown files one by one, without skipping any files
  • ✨ Adopt a progressive reading and summarization strategy, processing files in batches and managing context
  • ✨ Report generation requires segmented writing, ensuring content is within 2000 words and follows format specifications
  • ✨ All viewpoints must be supported by original text, prohibiting fabricated content
  • ✨ Citation links must use relative paths and link to specific files, with link text as titles
691 words · ~4 min read
  • AI Analysis Report
  • Document Processing
  • Markdown
  • Rule Guide
  • Automated Process
  • Summary Generation
  • Report Generation

Table of Contents

AI Analysis Report Generation - Basic RulesDocument Reading Process (Must be strictly followed)Core PrinciplePhase 1: Obtain Complete File ListPhase 2: Batch Reading and SummarizationPhase 3: Context ManagementPhase 4: Completeness VerificationPhase 5: Generate Report (Segmented Writing)Step 1: Write the Skeleton FileStep 2: Fill Content Section by SectionStep 3: Completeness CheckProhibited ActionsGeneral Rules1. Principle of Factual Basis2. Citation Link Standards3. Header Format4. Time Weighting

AI Analysis Report Generation - Basic Rules

Document Reading Process (Must be strictly followed)

Core Principle

This repository may contain hundreds or thousands of Markdown files. You must read each file one by one, without skipping any. To handle the large volume of files, adopt a progressive reading and summarization strategy.

Phase 1: Obtain Complete File List

  1. Execute npx czon@latest ls-files to get all Markdown files.
  2. Record the total number of files, N.
  3. Save the file list as a queue to be read.

Phase 2: Batch Reading and Summarization

Process files in batches. A batch size of 10-20 files is recommended, but you can adjust based on file size and complexity.

For each batch:

  1. Read: Use the Read tool to read the full content of each file in the batch, one by one.
  2. Extract: Extract relevant information based on the current style's focus points (see Style Guide below).
  3. Batch Summary: Organize the extracted information for that batch into a structured summary.
  4. Accumulate: Merge the batch summary into the "Knowledge Base" (see Knowledge Base structure in the Style Guide below).

Phase 3: Context Management

When the context limit is approached:

  1. Compress Knowledge Base: Compress the knowledge base according to the current style's compression strategy (see Style Guide below).
  2. Preserve Index: Regardless of compression, you must preserve the index of all file paths and titles.
  3. Continue Reading: Use the compressed knowledge base to continue processing the remaining files.

Phase 4: Completeness Verification

Before generating the report, you must confirm:

  1. All N files from the file list have been processed.
  2. The knowledge base contains basic information for each file (at least the path and title).

If omissions are found: Return to Phase 2 to process the missing files.

Phase 5: Generate Report (Segmented Writing)

Only after completing all the above phases can you begin generating the report.

Since the report content may be long, you must write it in segments. Writing the entire report at once is prohibited.

Step 1: Write the Skeleton File

First, use the Write tool to create the report file and write the following content:

  • Header format (Title, AI Analysis Time, Number of Files, Note, Separator Line)
  • Titles for all sections (titles only, no body text)
  • Place a placeholder marker below each section title: <!-- SECTION: [Section Name] -->

Example Skeleton:

# Report Title

**AI Analysis Time**: January 01, 2025
**Generated based on 42 Markdown files**
**Note**: This report is AI-generated. Content is for reference only.

---

## Overview

<!-- SECTION: Overview -->

## Second Chapter Title

<!-- SECTION: Second Chapter Title -->

...

Step 2: Fill Content Section by Section

In the order of the sections, use the Edit tool to replace each placeholder marker with the actual content, one at a time:

  1. Fill the content for only one section per operation.
  2. Replace <!-- SECTION: [Section Name] --> with the complete body text of that section.
  3. Limit the content written in a single operation to within 2000 words. If a section exceeds 2000 words, split it into multiple writes (first write the first part, leaving a temporary placeholder marker at the end, then continue writing the latter part).

Step 3: Completeness Check

After all sections are filled:

  1. Read the complete file to confirm there are no remaining <!-- SECTION: placeholder markers.
  2. Confirm all link formats are correct (relative paths start with ../, link text is the article title).
  3. Confirm the number of files N in the header is correct.

Prohibited Actions

  • ❌ Do not start generating the report before reading all files.
  • ❌ Do not skip any files, regardless of the total number.
  • ❌ Do not claim to "understand the whole picture" after reading only some files.
  • ❌ Do not fabricate non-existent files or content in the knowledge base.

General Rules

1. Principle of Factual Basis

  • Always base your work on factual content. Do not present information detached from facts.
  • All viewpoints must be supported by the original text.
  • Do not fabricate non-existent content.
  • When citing links to original text, ensure the links are valid.
  • Always link to specific Markdown files, not directories.
  • The link text should be the corresponding title, not the filename.
  • Since reports are generated in the .czon/AIGC/SUMMARY/ directory, use relative paths starting with ../../../ for references.

Correct Example:

[Capital Endurance War: The Strategy for Individual Investors to Transcend Class](../../../INSIGHTS/6.md)
[From Creation to Distribution—Building an AI-Native Content Engine](../../../INSIGHTS/4.md)

Incorrect Example:

[INSIGHTS/6.md](../../../INSIGHTS/6.md) ← Uses filename instead of title
[Capital Endurance War](../../../INSIGHTS/) ← Links to a directory

3. Header Format

Every report must begin with the following format:

# [Report Title]

**AI Analysis Time**: YYYY-MM-DD
**Generated based on N Markdown files**
**Note**: This report is AI-generated. Content is for reference only.

---

Where N is the actual total number of Markdown files obtained in Phase 1.

4. Time Weighting

  • Consider the time span, giving higher weight to more recent articles.
  • However, do not neglect important older content.

See Also