This project gives you research reports based on Claude Code source analysis. It focuses on how the app works, what it sends, what it can do, and how it manages prompts, tools, permissions, and context.
Use this repo if you want to:
- Read clear reports about Claude Code internals
- Learn how the agent loop works
- See how tool calls and MCP fit in
- Review system prompt behavior
- Understand telemetry and permission handling
- Explore persistence and hidden features
Visit this page to download:
If the page offers a release file, download it to your PC. If it gives you a folder or archive, save it and open it after the download finishes.
- Open the download page in your browser.
- Look for the latest release, file, or archive.
- Download the file to your Downloads folder.
- If the file is a ZIP, right-click it and choose Extract All.
- Open the extracted folder.
- Open the main report file in your browser or text app.
This repository is organized around deep research notes and source-based reports. The reports focus on the parts that matter most to end users and researchers.
Common topics include:
- Architecture
- Agent loop
- Tool system
- System prompts
- Telemetry
- Permissions
- MCP integration
- Context management
- Persistence
- Hidden features
You do not need coding skills to use this repo.
- Open the main page or report file.
- Read the sections one by one.
- Use the table of contents if the file includes one.
- Look for terms like “agent,” “prompt,” “tool,” and “telemetry.”
- Move between linked notes if the report includes them.
If you want a simple path, start with the architecture report, then read the agent loop section, then the tool system section.
This explains how Claude Code takes input, decides what to do, and moves through each step of a task. It helps you see how the system breaks work into smaller actions.
This section covers the tools Claude Code can call during a task. It may include file actions, command use, and other task helpers.
These reports look at the instructions built into the app. They show how the app shapes behavior before you type anything.
This part explains what data the app may collect or send. It helps you understand what happens in the background.
This section shows how access rules work. It covers what the app can ask for and when it needs approval.
MCP, or Model Context Protocol, lets the app connect to external tools and services. This report helps you understand how that link works.
This section explains how the app keeps track of what matters during a session. It may cover memory, limits, and what gets included in each step.
This covers how the app saves state or keeps data across runs. It helps you understand what stays and what resets.
🕵️ Hidden Features
This section points out behavior that may not show in the normal UI. It can include extra commands, checks, or internal paths.
This content is meant to be easy to open on a Windows PC.
Recommended setup:
- Windows 10 or Windows 11
- A modern browser such as Edge, Chrome, or Firefox
- At least 200 MB free disk space
- A stable internet connection for the download page
- A text viewer or browser that can open Markdown or HTML files
If the repo comes as a ZIP or folder, Windows Explorer is enough to open it.
If you download an archive, you may see files like these:
- README.md - Main entry file
- reports/ - Research notes and analysis pages
- notes/ - Shorter topic files
- assets/ - Images or linked files
- index.html - Simple file you can open in a browser
If you see a file named README.md, open it first. If you see index.html, double-click it to view the content in your browser.
This repo helps you get a clearer view of how Claude Code works under the hood. It can help you:
- Compare what the app shows with how it behaves
- Follow the path from input to action
- Learn how permissions affect each step
- See how tool use fits into the flow
- Understand what data may be stored or sent
The reports cover telemetry and system behavior so you can review how the app handles data. If you are checking how a tool works before using it, this repo gives you a way to read the details in one place.
Things you may want to look for:
- What is collected
- When data is sent
- How prompts are handled
- How long context stays active
- What triggers permission prompts
If you want the fastest route, use this order:
- Read the main architecture report.
- Read the agent loop section.
- Open the tool system notes.
- Check permissions.
- Review telemetry.
- Read MCP integration.
- Finish with persistence and hidden features
Use your browser search or file search for these terms:
- architecture
- agent loop
- tool
- prompt
- telemetry
- permission
- context
- persistence
- MCP
- hidden feature
- Download the ZIP from the link above.
- Find the file in your Downloads folder.
- Right-click it.
- Choose Extract All.
- Open the new folder.
- Open README.md or index.html
- Open the link above.
- Read the reports in the browser.
- Use the file links or section links to move between pages.
- Save the page if you want to keep a local copy.
- Markdown files for written reports
- HTML files for browser viewing
- JSON files for structured notes
- Text files for short findings
- Images for diagrams or references
This repository centers on:
- ai-agents
- anthropic
- claude-code
- code-analysis
- mcp
- reverse-engineering
- security-research
- source-analysis
- source-code
- system-prompts
- telemetry
If the repo has linked notes, use them in this order:
- Start with overview pages
- Move to architecture
- Open agent loop details
- Read tool system notes
- Check prompt behavior
- Review telemetry and permissions
- Finish with advanced and hidden feature notes
- Read one section at a time
- Use browser search for repeated terms
- Keep the download page open in another tab
- Use a large screen if you want side-by-side reading
- Save the folder after extraction so you can reopen it later