Dataset - Decoding the Configuration of AI Coding Agents: Insights from Claude Code Projects

data.json

It contains a collection of the 100 most popular repositories on GitHub, where for each repository, we have the CLAUDE.md files.

Root object containing two main properties:

collection_metadata (object): Aggregated statistics about the collection process

Collection date, method, and script version

Repository counts (analyzed, with/without CLAUDE files, failed)

Success metrics (rate percentage, total files found)

Redirect statistics (repositories and files using redirects)

repositories (array): List of 100 repository objects, each containing:

repository_info (object): GitHub metadata including name, URL, owner, description, stars, language, topics, dates, forks, and commits

collection_metadata (object): Per-repository collection details (timestamp, method, contributor count, files found)

claude_files (array): CLAUDE.md files discovered in the repository, each with:

File metadata: name, path, size, SHA hash

URLs: HTML view, API endpoint, raw download

Content: primary markdown content (content) and optional redirected content (redirected_content) for repositories using alternative documentation paths

Discovery metadata: method used, redirect flags, content lengths

claude_files_count (integer): Total files in repository

total_content_length (integer): Sum of all content lengths

RQ1.csv

Collection of titles ## extracted from data.json and their semantic categorization, and the URL of CLAUDE.md file from which this title was extracted.

Structure:

title (string): H2 section heading extracted from CLAUDE.md content

cluster_name (string): Semantic category assigned to the title (e.g., "Architecture", "Testing", "Development Guidelines", "Unclear")

url_claude (string): GitHub URL of the source CLAUDE.md file

RQ2.csv

Association of the type with the title and its categorization.

Structure:

cluster_name (string): Semantic category of the title.

title (string): Semicolon-separated list of H2 titles mapped to this category within the file

type (string): Comma-separated content types detected in the section (CODE, TEXT, LINK, DIAGRAM)

CODE: Fenced code blocks with explicit language identifiers

TEXT: Regular markdown text content

LINK: Markdown inline links [text](url)

DIAGRAM: Explicit mermaid diagram blocks

url_claude (string): GitHub URL of the source CLAUDE.md file

RQ3.csv

Result of FPMax associated with the data categorization.

Structure: 

clusters (string): Semicolon-separated list of co-occurring category names

size (integer): Number of categories in the co-occurrence pattern (2 or 3)

support (float): Proportion of the 328 total files containing this exact pattern (range: 0.0-1.0)

count (integer): Absolute number of CLAUDE.md files exhibiting this pattern

example_name (string): Example of section titles related to the extracted common patterns.