justxor/claudecourse

Claudecourse is a Russian-language written guide that explains how Claude Code, an AI coding agent, works under the hood. It is not a tool you install and run. It is a long educational document meant to teach the reader how modern coding agents are built, using Claude Code as the worked example. The guide walks through the ideas in layers, starting simple and getting more advanced. It opens by explaining what an agent actually is: unlike a plain chat model that only answers questions, an agent has tools it can use, such as reading files, running commands, and searching the internet, plus a loop that keeps taking actions and checking results until a task is finished. From there it covers the core agent loop, the tool system, permissions, slash commands, hooks and settings, and how external tools connect through something called MCP. Later sections move into more advanced territory: how the agent manages multiple sub-agents working together, how it keeps track of context as a conversation grows long, how it remembers information across sessions, and how sessions can be saved and restored. A final section covers production concerns such as observability, cost and token optimization, prompt caching, testing agents, security against prompt injection attacks, retry and timeout handling, and running things in parallel under load. The guide is organized so different readers can take different paths through it. A beginner is pointed toward the early setup and first-command sections. Someone who already uses these tools is pointed toward the practical recipes and best-practices sections. Someone who wants to build their own agent is pointed toward the architecture, sub-agent, and security sections. The author states the material is drawn from publicly available sources and community discussion, and is intended for study and technical research rather than as an official reference. The full README is longer than what was shown.