深习堂
课程价格指南博客

开放课程Skill DevelopmentAI Courses

Claude Code Best Practices

Set up clear project rules and settings, then verify that they work as intended

开始学习
全部章节
9
Locale
English
课程类型
开放课程

Organize your Claude Code project rules, settings, and permissions, and check which ones actually take effect. Choose extensions that suit routine tasks, then use loading records and logs to diagnose configuration problems step by step.

适合人群

  • Developers familiar with Claude Code basics who want to organize project rules and settings
  • Users who keep repeating project conventions or find that their settings do not take effect
  • Engineers defining team permissions and automation boundaries

解决的问题

  • Project conventions are scattered across chats and must be repeated in each new session
  • Settings override one another across files, and you cannot tell which values take effect
  • You do not know whether a routine workflow needs a Skill, command, or Hook
  • When rules or tools fail, you keep changing settings without knowing the cause

    • 学完收获

    • Write CLAUDE.md rules with a clear scope and requirements that can be checked
    • Work out how settings combine and test which actions are allowed, require approval, or are denied
    • Choose extensions for a routine workflow and define when they should be invoked
    • Define what evidence each stage of a complex task needs and when work should stop
    • Use a reproducible failure record to identify the cause, then repeat the original steps to verify the fix

    课程亮点

    • Review short configuration snippets and practice fixing specific mistakes
    • Check that a configuration exists, that it loads, and that it produces the intended behavior
    • Cover rules, settings, permissions, extensions, and troubleshooting without building a complete system at once

    全部章节

    9 个章节

    Lesson 1: Build Reliable Project Context with CLAUDE.md

    Select durable project guidance, place it at the right scope, rewrite vague rules as actionable requirements, and verify loading, conflicts, and behavior.

    Lesson 2: Manage How Claude Code Works with Settings

    Choose minimal Settings by sharing scope, derive effective values from sources, precedence, and field types, and verify why a setting did or did not take effect.

    Lesson 3: Control Claude Code's Operating Boundaries with Permission Rules

    Predict tool outcomes from rule sources and permission modes, narrow real-work authorization, and safely verify allow, ask, deny, and sandbox boundaries.

    Lesson 4: Package Reusable Ways of Working as Skills

    Design a reusable Skill with clear triggers, inputs, tools, supporting resources, and context boundaries, then verify discovery, execution, results, and failure stops.

    Lesson 5: Turn Common Operations into Slash Commands

    Identify each slash entry's real type, select stable repeatable workflows, design clear inputs, actions, outputs, and failure boundaries, and verify the loaded version.

    Lesson 6: Isolate Context and Parallelize Work with Subagents

    Decide when delegation repays its cost, provide a self-contained task with minimal capabilities, control parallel resources and context, and verify key claims, changes, and gaps.

    Lesson 7: Automate Notifications, Checks, and Safety Blocks with Hooks

    Design a well-bounded Hook for a concrete automation need, choose its event, match, and handler correctly, and verify control outcomes and failures with observable evidence.

    Lesson 8: Organize Complex Development Tasks into Verifiable Workflows

    Work backward from observable acceptance evidence, advance implementation in verifiable behavior slices, and set clear completion and stop boundaries for dependencies, parallel work, and loops.

    Lesson 9: Diagnose and Resolve Claude Code Problems

    Turn a vague anomaly into a reproducible incident, choose evidence along the runtime path, isolate the cause with one-variable controls, and verify the fix through the original failure path.