A practitioner's reference repo — ready-to-use templates for the files that make Claude Code work in the real world.
Companion to the Claude Code course on deeplearning.ai.
Templates and Config — copy these to your machine
| File/Folder | Where it goes on your machine |
|---|---|
CLAUDE_md_templates/ |
Copy chosen template to your project root as CLAUDE.md |
settings/settings.json |
~/.claude/settings.json (global) or .claude/settings.json (project) |
keybindings/keybindings.json |
~/.claude/keybindings.json |
hooks/ |
~/.claude/hooks/ (global) or .claude/hooks/ (project) |
slash_commands/ |
~/.claude/commands/ (global) or .claude/commands/ (project) |
Global vs. project: Files in
~/.claude/apply everywhere. Files in.claude/inside a project apply only to that project. Project settings override global ones.
Feature Guides — read these, then configure Claude Code
| Folder | What it covers | Relevant path on your machine |
|---|---|---|
channels/ |
Push external events into a running Claude Code session (Telegram, Discord, webhooks, CI) | Configured via --channels flag at session start |
dispatch/ |
Delegate tasks from your phone to your desktop — and trigger Claude Code via webhooks and external events | Configured via Claude Desktop + mobile app pairing |
scheduled_tasks/ |
Run recurring tasks on a schedule — CLI session-scoped (/loop) and durable Desktop tasks |
CLI tasks: session memory only. Desktop tasks: ~/.claude/scheduled-tasks/ |
remote_control/ |
Steer your local Claude Code session from your phone, tablet, or any browser | Configured via claude --remote-control or /remote-control |
memory/ |
How Claude remembers your project — CLAUDE.md vs. auto memory, memory hygiene, subagent memory | Auto memory: ~/.claude/projects/<project>/memory/ |
code_review/ |
Multi-agent PR analysis with severity-tagged findings + Auto-fix for live CI failures and review comments | GitHub App + admin setup at claude.ai/admin-settings/claude-code |
routines/ |
Cloud-based automation triggered by schedule, API call, or GitHub events — runs when your laptop is closed | Managed at claude.ai/code/routines or /schedule in the CLI |
agent_teams/ |
Coordinate multiple Claude Code instances with shared tasks and inter-agent messaging | Enable with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in settings.json |
managed_agents/ |
Managed Agents API orientation — multi-agent sessions, outcomes, dreams, and webhooks for production deployments | Anthropic API with managed-agents-2026-04-01 beta header |
enterprise/ |
Large codebase and enterprise deployment patterns — harness build sequence, CLAUDE.md at scale, skills and plugins, LSP integrations, maintenance and governance | Apply patterns in your project's .claude/ and root CLAUDE.md |
~/.claude/ # Global config (applies to all projects)
settings.json # Model, permissions, env vars
keybindings.json # Custom keyboard shortcuts
commands/ # Global slash commands
commit.md # Available as /commit everywhere
hooks/ # Global hooks
pre_tool_use_safety.py
projects/ # Auto memory (one directory per git repo)
<project>/memory/
MEMORY.md # Index — first 200 lines loaded every session
*.md # Topic files — read on demand
scheduled-tasks/ # Desktop scheduled tasks (Desktop app only)
<task-name>/SKILL.md
your-project/
CLAUDE.md # Project instructions for Claude
.claude/ # Project-level config
settings.json # Overrides global settings for this project
commands/ # Project-specific slash commands
hooks/ # Project-specific hooks
Configuration and Templates
- CLAUDE.md Templates — Instructions Claude reads before every session
- Settings — Model selection, permissions, environment variables
- Auto Mode — Auto-approve safe commands; reduce prompts with
/fewer-permission-prompts - Effort Levels — Tune Opus 4.7's adaptive thinking with
/effort - Recaps and Focus Mode — Summaries after long tasks; hide intermediate output with
/focus
- Auto Mode — Auto-approve safe commands; reduce prompts with
- Keybindings — Custom keyboard shortcuts
- Hooks — Scripts that run before/after tool use
- Slash Commands — Custom
/commandsfor repeated workflows- /go — Test, simplify, and open a PR in one shot
Feature Guides
- Channels — React to external events inside a running session
- Dispatch — Delegate tasks from your phone and trigger Claude Code via webhooks
- Scheduled Tasks — Automate recurring work on a schedule (local/Desktop)
- Remote Control — Access your local session from any device
- Memory — How Claude remembers your project across sessions
- Code Review + Auto-fix — Multi-agent PR analysis and automated CI/review response
- Routines — Cloud-based automation on schedule, API trigger, or GitHub events
- Agent Teams — Coordinate multiple Claude Code instances with the Advisor strategy (Opus/Sonnet/Haiku)
- Managed Agents API — Orientation to sessions, outcomes, dreams, and webhooks for production agent deployment
- Enterprise & Large Codebase Guide — Harness build sequence, CLAUDE.md at scale, skills and plugins, LSP integrations, and governance for large teams
Migrating from another AI coding tool? Anthropic's Import memory feature on claude.ai lets you export memory from ChatGPT, Gemini, or Copilot and bring it into Claude. For Claude Code specifically, the right home for that context is your
~/.claude/CLAUDE.md— your personal instructions file that applies across all projects. See Memory for how the two memory systems work.
Claude Code's power is mostly configuration, not code. The right CLAUDE.md prevents the most common mistakes. The right hooks enforce the guardrails your team needs. Slash commands turn your most-used workflows into one keystroke.
The features added in v1.1.0 extend this idea further. Channels, Scheduled Tasks, Remote Control, and Auto Memory don't change what Claude Code is — they change when and where it works. Claude Code is no longer only a tool you invoke. It can watch for events, run on a schedule, stay reachable from your phone, and learn from your corrections over time.
Dispatch completes that picture. With Dispatch, you don't need to be at your desk to assign work. Send a task from your phone, walk away, and come back to results. Pair it with Channels for webhook and chat app triggers, and Claude Code becomes something closer to a background agent than a coding assistant — one you can reach from anywhere and hand off work to at any hour. The configuration layer got bigger. The principle stayed the same.
Opus 4.7 shifts the balance further toward autonomy. Auto mode replaces permission babysitting with a model-based safety classifier. Effort levels replace manual thinking-budget tuning with a single dial. Focus mode and Recaps let you step away from long tasks and return to a clean summary. The /go skill closes the loop: test, simplify, ship. The principle is the same as always — configuration over code — but the ceiling on what you can hand off keeps rising.
v2.0.0 extends the field guide into the GitHub loop and the cloud. Code Review and Auto-fix bring Claude into the PR lifecycle — not just writing code but analyzing it before merge and watching it after, responding to CI failures and reviewer comments autonomously. Routines push the "works when your laptop is closed" idea to its logical conclusion: fully cloud-based automation triggered by a schedule, an API call, or a GitHub event, with no local process required. Agent Teams introduce collaborative multi-agent work with the Advisor strategy — Opus at the top making judgment calls, Sonnet handling implementation, Haiku handling tools — a tiered model configuration that balances capability and cost. And the Managed Agents API section orients practitioners toward Anthropic's production-grade agent layer, where sessions, outcomes, dreams, and webhooks give you programmatic control over agents you're building into products rather than using yourself.
The ceiling keeps rising. The principle stays the same.
v2.1.0 brings the field guide to enterprise scale. The central thesis, drawn from Anthropic's Applied AI team's observations across large deployments, is that the harness matters more than the model. Two teams with identical model access but different harness configurations produce dramatically different results. The enterprise section makes this concrete: a build sequence that follows the model's own behavior (CLAUDE.md before hooks, skills before plugins, connectors last), layered CLAUDE.md hierarchies that load the right context without bloating every session, skills and plugins that turn tribal knowledge into distributed infrastructure, and LSP integrations that give Claude symbol-level navigation where text search fails at scale. At enterprise scale, configuration becomes organizational — the right harness isn't just files you write, it's a living system a team owns, reviews on a cadence, and governs deliberately. Cross-functional working groups, designated ownership, and AI code review policies aren't bureaucracy. They're the harness applied to the human layer. The ceiling keeps rising. The principle stays the same.
v2.1.0 2026-05-17 Written by Claude Code in collaboration with ArchieCur and Sonnet 4.6 Enterprise section developed from Anthropic Applied AI team, "How Claude Code works in large codebases" (May 2026)