Imported from AI-agents-incubator/essays (code-agents/claude-code-book/AGENTS.md). Install upstream with npx skills add AI-agents-incubator/essays --skill claude-code-book. Copyright stays with the author.

Claude Code Book Instructions

Scope

These instructions apply to:

• code-agents/claude-code-book

Goal

Build a full, self-sufficient book on Claude Code for non-technical readers. This is not a developer manual. This is not a terse outline. This is not a glossary with fragments. Each chapter should read like part of a practical book.

Audience

• Non-programmers
• Domain experts
• People who may freeze on every unfamiliar interface detail

Core Writing Standard

Whenever the text says that the user should do something, explain:

• where they are on the screen
• what they should see
• what button or UI element to look for
• what may be written on that button
• what should happen after clicking
• what to do if they do not see the expected thing

Chapter Structure

Each chapter should include:

• who the chapter is for
• what the reader will be able to do after it
• conceptual explanation in simple language
• detailed practical steps
• bad and good examples
• common mistakes
• a practical exercise
• glossary
• reference links only as supporting material

Workflow Rules

• One chapter = one file.
• After each new chapter, update:
  • code-agents/claude-code-book/README.md
  • root README.md
• Keep the chapter self-sufficient.
• Do not require the reader to leave the chapter to understand the chapter.

Style Rules

• Plain Russian
• Calm, precise, non-hype tone
• No unnecessary developer slang
• No "just do X" without interface explanation
• No hidden assumptions about Git, terminal, IDEs, or repo structure

Autonomy Rules

• Continue through P1 backlog items without asking after every chapter.
• Pause only for editorial forks, scope changes, or uncertainty that would change the book's structure.

Pipeline Model Selection

• When a pipeline/API model choice arises, consult GPT-5.6 Model Selection Guide, map its Model × Reasoning recommendation to the provider/model supported by this project, and do not default to mini/nano by habit.
• For a cross-provider audit or model-refresh recommendation, read Pipeline Model-Selection Handoff, ground it in live code and current official provider documentation, and return a read-only evaluation plan before changing any runtime route.

Autonomous Orchestration

This project inherits the global Codex orchestration policy from /Users/alexeykrolmini/.codex/AGENTS.md.

• Codex is the orchestrator and integrator for work inside the approved scope: it owns decomposition, session/sub-agent creation, supervision, evidence review, integration, verification, and session lifecycle.
• Create parallel or successor sessions whenever independent ownership, worktree isolation, assurance, long-running work, recovery, or context separation makes them useful. Transcript capacity is only one trigger.
• Use /Users/alexeykrolmini/.codex/skills/handoff/SKILL.md for task charters, continuations, recovery, and silent control-plane rollover. The user is not responsible for copying prompts or coordinating sessions.
• Preserve one accountable controller/integrator, one writer for shared contracts, unrelated dirty-tree work, and truthful evidence states.
• Stop for explicit Product Owner approval before any production mutation or spending action unless an applicable strict pre-approved strategy states the exact targets, limits, expiry, evidence, rollback, and remaining stop gates.
• Project rules may narrow authority or add checks; they do not transfer routine orchestration and integration mechanics back to the user.