claude-code-guide: A Comprehensive Community Reference for Claude Code

Claude Code ships with official documentation, but the docs are organised around concepts, not around the question you have at 2pm when the API key rotation just broke your build. zebbern/claude-code-guide is a community-curated reference structured the way you'd actually look things up — install paths, environment variables, slash commands, settings precedence, Plan/Auto mode, hooks, MCP, security, automation recipes, troubleshooting, and third-party integrations.

Why This Skill Matters

The guide is organised into nine top-level sections that mirror how someone setting up Claude Code actually moves through a problem: Getting Started, Configuration & Environment, Commands & Usage, Interface & Input, Advanced Features, Security & Permissions, Automation & Integration, Help & Troubleshooting, and Third-Party Integrations. Each section is dense with cheat-sheet tables and concrete examples rather than narrative paragraphs.

Concrete things covered: the official native installer (curl -fsSL https://claude.ai/install.sh | bash), Homebrew, npm, and package-manager variants; the full environment-variable surface (auth, routing, model aliases, Bedrock/Vertex/Foundry providers, timeouts, session controls, network routing, privacy flags); configuration-file precedence (managed > CLI > local > project > user); slash commands, keyboard shortcuts, vim mode, Plan Mode, Auto Mode, Background Tasks; hooks; sub-agents (.claude/agents/<name>.md); Skills (SKILL.md layout); PR-review and issue-triage automation recipes; MCP server scope and recommendations; troubleshooting (common errors, debugging techniques, paths, proxies); DeepSeek integration and provider setup examples.

This is a reference document, not an installable skill. The README explicitly notes that commands and provider model mappings change quickly and links to the official Claude Code documentation as authoritative. License is MIT. Stars 4.6k, forks 462.

Installation

There is nothing to install. Open the README, browse the table of contents, and use the fast-path links (Install, Commands, Config, MCP, Agents, Troubleshoot) at the top.

git clone https://github.com/zebbern/claude-code-guide

Use it like a long-form cheat sheet.

Real Workflow: Audit a Stuck Claude Code Session

You are stuck because something just stopped working. The guide's troubleshooting section is the fastest starting point.

Step 1. Open the README and jump to Help & Troubleshooting.

Step 2. The Common Errors and Solutions table covers the most frequent failures (auth rotation, sandbox permissions, MCP server crashes). Find the closest match to your symptom and try its fix.

Step 3. If the table does not have your case, the Debugging Techniques section runs through the layered debugging flow: enable verbose logging, inspect the MCP server logs, check environment-variable precedence (managed > CLI > local > project > user), confirm the model alias resolves to a provider you have access to.

Step 4. For proxy, or firewall issues, the Paths and Proxies section lists the directories Claude Code reads on each OS plus the environment variables that route network traffic.

Step 5. If the issue is provider-specific (DeepSeek, Bedrock, Vertex, Foundry), the Third-Party Integrations section has setup examples and known gotchas per provider.

Real Workflow: Build a Workflow From Setup to PR Review

You are setting up Claude Code for a new project and want a checklist that takes you from zero to a working automated PR review on GitHub. The guide lays it out section by section.

Step 1. Getting Started → run the native installer (curl -fsSL https://claude.ai/install.sh | bash). Configure completion alerts via /config and pick a notification channel.

Step 2. Configuration & Environment → set the auth and routing variables for your provider. Confirm with the cheat sheet's Config essentials block that the precedence order matches what you expect.

Step 3. Commands & Usage → pick the slash commands you will rely on day-to-day. Pin them in a project-level AGENTS.md so every session inherits the conventions.

Step 4. Advanced Features → enable Plan Mode for changes that touch multiple files, Auto Mode for single-file edits, Background Tasks for long-running shell work. Set up Sub-Agents (.claude/agents/<name>.md) for any workflow you repeat more than three times a week.

Step 5. Automation & Integration → copy the PR-review recipe, point it at your repo's CI trigger, and verify the auto-suggested labels and severity tiers match your team's conventions.

Step 6. Security & Permissions → set the dangerous-mode flag explicitly off, configure permission scopes for any tools your agents use (especially MCP servers and Bash), and review the agent skill files before installing.

Step 7. Help & Troubleshooting → bookmark the Paths and Proxies table; you will need it the first time a corporate firewall breaks the install.

Real Workflow: Triage a Skill You're Considering

You are evaluating a third-party skill (one you found on this very site, for example) and want to know what to verify before installing. The guide's Security & Permissions and Skills sections together cover the checklist.

Step 1. Read the Skills section to confirm the SKILL.md layout the guide documents matches the one your candidate ships with. Mismatches are a red flag — the skill is probably hand-rolled and not following the Agent Skills open standard.

Step 2. Confirm the skill declares its trigger conditions in the frontmatter (description field) and in a When to use / When not to use body section. Skills without these make it impossible to scope.

Step 3. Cross-check the install path against the guide's documented install paths. Mismatches here mean the skill is not in the standard npx skills or /plugin marketplace ecosystem and likely cannot be uninstalled cleanly.

Step 4. If the skill uses hooks (the guide covers this in the Advanced Features section), audit the hook for any file-rewrite or network-call side effects before installing. Hooks are the most common foot-gun in third-party skills.

Step 5. Run the skill in a sandbox (a throwaway branch or a worktree). The guide covers worktree-based isolation in the Automation section. Treat the throwaway run as a security gate, not just a smoke test.

Tips

• Use the table-of-contents fast paths (Install, Commands, Config, MCP, Agents, Troubleshoot) when you land on the README cold.
• Treat the guide as a lookup, not a tutorial — the official Claude Code docs remain authoritative, especially for commands and provider mappings that change quickly.
• Pin a working .claude/agents/<name>.md and AGENTS.md to your repo so the conventions you read in the guide survive across sessions.
• For security-sensitive installs, run the skill in an isolated worktree first; check the hooks section before approving any auto-rewrite behaviour.
• Provider integrations (Bedrock, Vertex, Foundry, DeepSeek) each have a setup example in the Third-Party Integrations section — follow them rather than improvising.

When Not to Use This

If you are looking for a runtime installable skill, this is a reference document, not one. Pick a specific skill from the catalog (such as antfu/skills, heilcheng/awesome-agent-skills, or the project-specific skills this site covers) for installation. Also, since commands and provider mappings change quickly, the linked official Claude Code documentation remains the source of truth — treat this guide as a fast lookup rather than a definitive reference.

See the leaderboard for more guides and references.