A practical guide, architecture manual, and operator playbook compiled from Anthropic's public workshops, engineering handbooks, and power user workflows.
Filter by focus area to master prompt caching, custom skills, deterministic Stop hooks, and enterprise PR automation.
Install the native binary, complete OAuth terminal login, run your first codebase query, edit safely, and manage git commits without leaving the shell.
Deep dive into the 3-step loop (gather context, take action, verify results), 5 built-in tool categories, rewind checkpoints, and permission security modes.
Understand how prefix matching keeps sessions fast and cost-efficient, what actions invalidate the cache, and how to configure 5-minute vs 1-hour TTLs.
Battle-tested prompt recipes for navigating unfamiliar repos, reproducing bugs with stack traces, safe refactoring, and auto-scaffolding edge-case tests.
Distinguish deterministic harness enforcement from non-deterministic model prompts. Keep CLAUDE.md lean (<200 lines) and enforce quality gates with Stop hooks.
Select the optimal surface for your workflow across terminal CLI, Desktop GUI, VS Code, JetBrains, Web, Mobile Dispatch, and team Slack channels.
Set up multi-agent PR reviews on GitHub. Claude reads diffs in parallel, verifies findings against the codebase, filters false positives, and posts inline comments.
Run concurrent feature implementations, bug fixes, and reviews on separate checkouts without file collision using claude --worktree.
Connect Claude to databases, issue trackers, and third-party APIs via Model Context Protocol (MCP) and author custom skills in .claude/skills/.
Run Claude non-interactively in shell loops, git hooks, and GitHub Actions with --bare, --allowedTools, and JSON streaming.