Deep-dive documentation for AI CLI tools — the stuff official docs don't cover.
For developers, researchers, and power users who need to understand how these tools actually work under the hood. We extract undocumented features, hidden APIs, architecture internals, gotchas, and real implementation patterns directly from source code.
Why use these docs? Official docs tell you what to do. We show you how it works, what's not documented, and what breaks.
All content extracted from public open-source repositories — no proprietary info, no secrets, just better organized knowledge.
Open source doesn't automatically mean well-documented. While these projects are publicly available, critical details often remain buried in implementation code:
- Undocumented features — Experimental APIs, beta functionality, and hidden capabilities not in official docs
- Implementation internals — How things actually work vs. what the public API exposes
- System prompts — The actual instructions sent to LLMs that shape agent behavior
- Real-world patterns — Gotchas, edge cases, and integration nuances learned from source code
This repository started when I tried to understand how the Task tool works under the hood across different CLI agents. That single question led to extracting architecture diagrams, security models, tool systems, and undocumented flows that weren't explained anywhere else.
For developers and agents: Use this as a reference when building AI workflows, understanding tool implementations, or debugging complex behaviors. These docs bridge the gap between "read the source" and "trust the marketing page."
TypeScript library by Anthropic for building AI agents powered by Claude. 17 built-in tools, MCP integration, security hooks, specialized sub-agents, and flexible permission modes.
Google's terminal CLI for Gemini AI. 5-package system with 15+ tools, agent-to-agent communication, VS Code integration, and MCP support.
OpenAI's Rust-based terminal coding agent. High-performance with advanced tools, security sandboxing, and comprehensive approval policies.
Open-source, provider-agnostic AI coding agent built for the terminal. Client/server architecture with built-in LSP, supports multiple AI providers (Anthropic, OpenAI, Google, Bedrock), and features ACP/MCP protocol integration.
Smart AI code and GitHub research MCP server for exploring this repository. Provides advanced pattern matching, semantic search, and cross-reference analysis across all documentation.
→ github.com/bgauryy/octocode-mcp | Demo
claude_code_docs.mp4
Know an AI/CLI tool with incomplete docs, hidden features, or complex internals? Open an issue with the repo URL and what's missing.
Have specific questions about documented repositories? Also welcome! Open an issue with your questions about architecture, implementation details, or undocumented behaviors.
Star on GitHub • Share with others • Request documentation for your projects • Sponsorships help maintain this project
Research and educational purposes only. Content derived from public open-source resources. Verify all information independently. No warranty provided. Use at your own risk.
Enterprise Documentation: Need enterprise-level documentation for your organization's codebase? Reach out: [email protected]