How AI Coding Agents Learn Your Repo: Codex, Claude Code, Gemini CLI скачать в хорошем качестве

How AI Coding Agents Learn Your Repo: Codex, Claude Code, Gemini CLI

How AI Coding Agents Learn Your Repo: Codex, Claude Code, Gemini CLI This video is a practical architecture deep dive into the instruction layer behind modern AI coding tools. Instead of repeating the broad CLI overview, it zooms in on one question: how does a coding agent become repo-aware in a way that is reliable enough to trust? Topics covered: repo-local instruction files such as `AGENTS.md`, `CLAUDE.md`, and `GEMINI.md` how hooks, skills, and command surfaces keep guidance lazy instead of bloating every turn why MCP matters as the capability bus for tools and services why trust, approval, and sandboxing are separate design layers how context assembly differs from durable repo instructions how to structure a codebase so agents act more predictably Primary sources are official docs current as of March 11, 2026. Primary sources: OpenAI Codex CLI: https://developers.openai.com/codex/cli/ OpenAI AGENTS.md: https://developers.openai.com/codex/g... OpenAI Skills: https://developers.openai.com/codex/g... OpenAI MCP: https://developers.openai.com/codex/g... OpenAI Sandboxing: https://developers.openai.com/codex/c... Anthropic Claude Code overview: https://code.claude.com/docs/en/overview Anthropic Hooks: https://code.claude.com/docs/en/hooks Anthropic MCP: https://code.claude.com/docs/en/mcp Gemini CLI docs: https://geminicli.com/docs/cli/ GEMINI.md: https://geminicli.com/docs/core/gemin... Gemini trusted folders: https://geminicli.com/docs/trusted-fo... Gemini policy engine: https://geminicli.com/docs/reference/... If you build local AI tools, the lesson is that repo awareness is not one prompt. It is a control plane. #ai #codex #claudecode #geminicli #mcp #softwarearchitecture Chapters 0:00 How AI Coding Agents Learn Your Repo 0:40 From CLI Architecture To Repo Behavior 1:26 Repo Awareness Is Loaded, Not Guessed 2:20 Instruction Precedence 3:16 Path-Local Instruction Files 4:10 What Good Instructions Contain 5:04 Hooks, Skills, And Commands 6:14 MCP And The Tool Runtime 7:10 Trust, Approval, And Sandboxing 8:16 Context, Memory, And Checkpoints 9:28 Designing An Agent-Friendly Repo 10:34 Failure Modes That Break Trust 11:20 Reference Architecture 12:10 The Takeaway