--- name: claude-code-guide description: "Claude Code guide skill: CLI workflows, project context, tool permissions, slash commands, hooks, MCP, large-file analysis, debugging, and advanced coding-agent operating patterns." --- # claude-code-guide Skill Use this skill to answer Claude Code workflow questions and to turn the long Chinese guide in `references/README.md` into focused operating steps. ## When to Use This Skill Trigger when any of these applies: - Learning or explaining Claude Code CLI workflows, command habits, or project-context handling. - Designing slash commands, hooks, MCP integrations, memory/context conventions, or task workflows. - Debugging Claude Code permission, tool-use, file analysis, or large-context problems. - Comparing interactive coding-agent patterns such as explore-plan-edit-verify, review loops, or multi-agent coordination. - Extracting a compact answer from the full Claude Code guide in `references/README.md`. ## Not For / Boundaries - Not the source of truth for current Claude product availability, pricing, or model names; verify those with official Anthropic docs when exact current behavior matters. - Not for bypassing tool permissions, sandbox rules, repository policies, or higher-priority system/developer instructions. - Do not treat the long guide as automatically correct for every Claude Code version; use it as local project reference material. - Required inputs: target workflow, OS/shell, Claude Code version if relevant, current repository constraints, and the exact error or behavior observed. - If a command may mutate files or external systems, identify the blast radius and validation path before recommending it. ## Quick Reference ### Common Patterns **Start with project context** ```text Read README/AGENTS/CONTRIBUTING, inspect git status, then search for local patterns before editing. ``` **Large-file triage** ```bash wc -l path/to/file.md rg -n '^#{1,6} ' path/to/file.md sed -n '1,120p' path/to/file.md ``` **Slash command authoring shape** ```text .claude/commands/.md ``` **Hook design checklist** ```text event -> allowed command -> timeout -> logging -> rollback/disable path ``` **MCP integration checklist** ```text server name -> command/env -> auth source -> minimal permission -> smoke test ``` **Debug a Claude Code failure** ```text Capture: command, working directory, permission mode, exact error, relevant config, and smallest reproduction. ``` **Context hygiene rule** ```text Keep entry prompts short; move long references to files and load only the needed section. ``` ## Examples ### Example 1: Add a Custom Slash Command - Input: user wants a repeatable `/review-pr` workflow. - Steps: 1. Create or update `.claude/commands/review-pr.md` with scope, required inputs, and output format. 2. Reference repository review rules instead of duplicating stale policy text. 3. Test the command on a small diff and refine ambiguous steps. - Expected output / acceptance: command is discoverable, deterministic, and does not bypass repo review constraints. ### Example 2: Diagnose a Permission Problem - Input: Claude Code refuses to run or edit a file. - Steps: 1. Identify whether the blocker is tool permission, filesystem sandbox, repository policy, or OS permissions. 2. Capture the exact path, command, and error. 3. Propose the least-permissive fix and a smoke test. - Expected output / acceptance: user gets a concrete unblock path without granting broad unnecessary access. ### Example 3: Summarize a Long Guide Section - Input: need the MCP section from `references/README.md`. - Steps: 1. Use heading search to locate the section. 2. Read only the relevant slice. 3. Return a short operator checklist plus links back to the local reference file. - Expected output / acceptance: answer is actionable and avoids dumping the full long document. ## References - `references/index.md`: structured navigation for the guide. - `references/README.md`: long-form Claude Code Chinese guide. ## Maintenance - Sources: local `references/README.md` and `references/index.md`. - Last updated: 2026-04-28 - Known limits: Claude Code behavior can change; verify current CLI flags, hooks, and MCP details against official docs when precision matters.