llms.txt

# Claude Code Ultimate Guide

> The most comprehensive documentation for Claude Code CLI by Anthropic. Covers beginner to advanced: hooks, skills, agents, MCP servers, context engineering, memory systems, multi-agent orchestration, and production patterns.

Guide version 3.43.0, tracking Claude Code up to v2.1.250 (August 2026).

## What is covered

- Getting started: installation, CLAUDE.md setup, vocabulary, first session, 7-module learning path (8-11h)
- Executable learning: installable learning-path skill with prerequisite gates, local evidence, scheduled reviews, and a tested seven-module companion project
- Context window management: 1M-token context on Sonnet 5 and Opus 5, context rot, /compact and /clear thresholds, auto-compact
- Context engineering: managing what the model sees, when and in what shape; token audit workflow, compression tools, ejection discipline, maturity model
- Memory systems: native stack (CLAUDE.md levels, auto memory, auto dream, agent frontmatter), cross-session tools (claude-mem, agentmemory, ICM, Kairn, OpenMemory), team sharing, file-based experience playbooks, memory poisoning risks
- Hook system: all 30 hook events with matchers, input schemas, exit-code-2 behavior and decision-control formats; blocking vs non-blocking
- Event ingestion and safe delegation: Monitor command/WebSocket sources, plugin monitors, MCP Channels, Routines GitHub triggers, and a read-only-first Codex workflow
- Slash commands: full categorized table of the ~100 built-in commands with aliases, mirroring the official reference at https://code.claude.com/docs/en/commands (the source of truth), plus removed and renamed commands
- Skills and slash commands: authoring, trigger patterns, progressive disclosure, skill design patterns
- Current product surfaces: Computer Use, plugin packaging and recommendation hints, Claude apps gateway, artifact publishing
- Agents: subagent frontmatter, delegation, agent teams, agent evaluation metrics, harness engineering
- Agent harnesses: model-harness pair evidence, explicit loop and graph contracts, judgment allocation, durable execution, reviewer-independence tests, four operating layers plus an outer optimizer layer, a pinned 160-project source directory plus 32 guide supplements, 42 strict runtimes, 15 adjacent control planes, a research meta-harness table, a test-drive protocol, and DeepSeek Harness plus Liza profiles
- Multi-agent orchestration: parallel agents, hub-and-spoke coordinators, dynamic workflows (JavaScript-orchestrated pipelines, `ultracode`), agentic software factories, AI executive agents (virtual C-suites and board simulators applying the persona/agent pattern to named business roles)
- MCP server integration: setup, custom servers, validated community server catalog, MCP vs CLI vs Skills trade-offs, advanced tool design
- Guide MCP product reference: guide/ecosystem/claude-code-guide-mcp.md covers client installation, published versus candidate capabilities, architecture, privacy, offline behavior, troubleshooting, and dated npm/GSC/GA4 metrics
- Security and permission model: .claude/settings.json, permission rules, sandbox isolation, threat database, data privacy and retention, enterprise governance (guardrail tiers, MCP approval workflow, audit trails)
- CI/CD and code review: GitHub Actions patterns, the Code Review feature (Teams/Enterprise), multi-provider review pipelines
- Verification methods: bounded Best-of-N selection, independent review limits, executable checks, and TESTING.md proof records
- Ops and cost: observability and session forensics, API gateway for enterprise cost control (LiteLLM, Portkey), AI unit economics, subscription strategy at team scale (Claude, Codex, Copilot, Gemini, Cursor, Mistral, governed API traffic, and self-hosted inference), team metrics (DORA, SPACE, agentic metrics)
- Team knowledge infrastructure: 3-tier framework (Markdown vault, MCP connectors, RAG at scale), Code+Cowork bridge, Atlassian/Notion/GitBook MCP, Onyx/LlamaCloud/Ragie
- Adoption: L0-L5 maturity scale, the J-curve, empirical data (METR RCT, DeputyDev cohort), roles in AI-assisted teams
- Practitioner insights: condensed findings from IFTTD podcast episodes, Devoxx, ByteByteGo, Stanford Online, plus 150+ scored external resource evaluations
- Production templates: ready-to-use CLAUDE.md, rules, hooks, agents, skills, commands, quizzes, GitHub Actions workflows

## Machine-readable entry points

- machine-readable/reference.yaml: full index with file paths, anchors and line numbers into the guide
- machine-readable/claude-code-releases.yaml: condensed Claude Code release history, breaking changes, milestones
- machine-readable/cowork-reference.yaml: index for Claude Cowork (dedicated repo, non-dev audience)
- machine-readable/agentsec-security-feed.v1.json: AgentSec versions, detector coverage, security counters, and reviewed incident fiches consumed by the landing
- machine-readable/agent-harnesses.json: normalized Agent Harness Map with evidence states, official project URLs, and dated repository metadata
- machine-readable/mcp-product.json, mcp-stats.json, and mcp-dashboard.json: generated runtime contract plus dated package and aggregate page metrics with explicit availability and methodology boundaries
- machine-readable/translations.json: canonical, maintained, and community translation status with source checksums, versions, dates, and coverage
- machine-readable/distribution-channels.yaml: attributed publication channels, asset states, dates, and 30-day outcome fields

## Key URLs

- Guide site: https://cc.bruniaux.com
- Guide reader: https://cc.bruniaux.com/guide/
- Agent Harness Engineering: https://cc.bruniaux.com/guide/agent-harness/
- Loop & Graph Engineering: https://cc.bruniaux.com/guide/loop-graph-engineering/
- Agent Harness Map: https://cc.bruniaux.com/guide/agent-harness-landscape/
- Agent Tools: https://cc.bruniaux.com/guide/agentic-tools/
- Glossary: https://cc.bruniaux.com/guide/glossary/
- Translations and language status: https://cc.bruniaux.com/guide/translations/
- Learning path skill: https://github.com/FlorianBruniaux/claude-code-ultimate-guide/tree/main/examples/skills/learning-path
- Learning companion project: https://github.com/FlorianBruniaux/claude-code-ultimate-guide/tree/main/examples/learning-project
- Best-of-N workflow: https://github.com/FlorianBruniaux/claude-code-ultimate-guide/blob/main/guide/workflows/best-of-n.md
- Releases tracking: https://cc.bruniaux.com/releases/
- Whitepapers (PDF + EPUB, FR and EN): https://cc.bruniaux.com/whitepapers/
- GitHub: https://github.com/FlorianBruniaux/claude-code-ultimate-guide
- Plugins marketplace: https://github.com/FlorianBruniaux/claude-code-plugins
- Cowork guide (non-devs, Claude Desktop): https://github.com/FlorianBruniaux/claude-cowork-guide

## Author

Florian Bruniaux, AI engineer with 12+ months of production Claude Code usage. Former VP Engineering. Expert in context engineering and AI-assisted development workflows.

- Portfolio and blog: https://florian.bruniaux.com
- LinkedIn: https://www.linkedin.com/in/florian-bruniaux-43408b83/