模块 1: 基础1.7: Claude Code 导航

1.7: Claude Code 导航

  • 完成时间： 15-20 分钟
  • 前提要求： 完成模块 1.1-1.6

在 Claude Code 中启动这个模块： 运行 /start-1-7 来开始互动体验。

📖 概述

模块 1.7 涵盖最终的导航要点：输入模式、分析深度控制和工作流加速。这些技能将普通用户与高级用户区分开来。

你将学到：

  1. 输入模式 - 编辑、自动接受和计划模式
  2. 思考关键词 - 控制分析深度（think/think harder/ultrathink）
  3. —dangerously-skip-permissions - 关键工作流加速器

关键要点： 掌握何时使用每种模式，根据需求调整分析深度，并从工作流中消除摩擦。

🎛️ 三种输入模式

Claude Code 有三种输入模式来控制工作流执行：

模式行为最适合切换方法
编辑（默认）显示差异，等待批准关键文件、学习、安全按 Shift+Tab
自动接受立即应用更改简单任务、可信操作按 Shift+Tab
计划首先创建带待办事项的结构化计划多步骤工作流、复杂研究按 Shift+Tab

切换： 随时按 Shift+Tab 在模式间循环。

编辑模式（默认）

在应用之前显示每个文件更改。你审查并批准每个更改。

何时使用：

  • 处理关键文件
  • 学习 Claude 如何处理任务
  • 需要控制和审查

自动接受模式

自动应用更改而无需询问。快速且自主。

何时使用：

  • 任务简单明了
  • 信任 Claude 正确执行
  • 想要速度而不是审查

计划模式

在执行之前创建带有自动生成待办事项列表的结构化计划。提前了解策略。

Use when:

  • Complex, multi-step workflows (3+ distinct steps)
  • Want to see strategy before execution
  • Research → synthesis → deliverable workflows
  • Multiple agents need coordination

Example todo list:

☐ Research SingTech's main competitors
☐ Launch parallel agents to analyze AI features
☐ Synthesize findings into competitive analysis
☐ Create response strategy document

As Claude works:

✓ Research SingTech's main competitors
→ Launch parallel agents to analyze AI features
☐ Synthesize findings into competitive analysis
☐ Create response strategy document

🎯 Mode Selection Guide

TaskRecommended ModeWhy
Research 5 competitors and create strategyPlan ModeComplex, multi-step, benefits from visibility
Add dark mode toggleAuto-Accept/EditSimple, one-step task
Fix typo in READMEAuto-Accept/EditSimple, low-risk
Create PRD from rough notesPlan ModeMulti-step: research → outline → draft
Analyze CSV fileAuto-AcceptStraightforward analysis
Multi-source research synthesisPlan ModeComplex: agents → synthesis

Quick Decision Tree:

  1. Complex multi-step workflow? → Plan Mode
  2. Straightforward and low-risk? → Auto-Accept
  3. Otherwise → Edit Mode (safest)

💡 Think Control Keywords

Adjust analysis depth before Claude responds:

KeywordDepthWhen to Use
think about XNormalMost tasks
think harder about XDeeperComplex decisions
ultrathink about XMaximumReally hard problems (displays as rainbow!)

Examples:

With plan mode:

Think harder about competitive threats, then research competitors' AI features
and create response strategy

With any mode:

Ultrathink about whether we should build or buy this feature

Pro tip: Think keywords work in all input modes, especially useful in plan mode for complex strategy work.

⚡ Dangerously Skip Permissions

The Flag

claude --dangerously-skip-permissions

What it does: Skips all permission prompts. Claude executes commands without approval.

The transformation:

Without FlagWith Flag
Can I write to user-story-1.md? → Yes
Can I write to user-story-2.md? → Yes
Can I write to user-story-3.md? → Yes
[Repeat for every file…]		[Creates all 5 files immediately]
Done!

Why It’s Worth It

The tradeoff:

  • Risk: No safety net, commands execute immediately
  • Reward: Pure flow state, 10x faster, real workflow speed

Carl’s Strong Recommendation

Use it once you’ve completed Level 1 and trust Claude Code. The speed gains are worth it.

Create a shell alias:

alias cc="claude --dangerously-skip-permissions"

Then just type cc to start Claude Code in fast mode.

When to use:

  • Completed Level 1
  • Trust Claude Code
  • Working in safe environment
  • Want professional speed

When to avoid:

  • Production code
  • Unsure what Claude will do
  • Learning new project

🛑 Stopping Execution

Press Esc at any time to stop Claude from executing.

What happens:

  • Execution stops immediately
  • Files created so far remain
  • Can review and decide next steps

Pro tip: In plan mode, press Esc once you’ve seen the value, then review files when ready.

💡 Best Practices

Do:

  • Use plan mode for multi-step workflows (3+ steps)
  • Switch modes based on task complexity
  • Use think keywords for deeper analysis on complex decisions

Don’t:

  • Use plan mode for simple tasks (overkill)
  • Stay in one mode all the time (be flexible)
  • Forget Esc stops execution

Workflow Development:

  • Week 1: Edit mode for everything (learn how Claude works)
  • Week 2: Auto-accept for simple tasks
  • Week 3: Plan mode for complex workflows
  • Week 4: Natural mode switching based on task

❓ Common Questions

When should I use plan mode vs just asking Claude to do it?

Use plan mode when task has 3+ distinct steps or involves research → synthesis → deliverable. Skip for simple, obvious tasks.

Can I edit the plan before Claude executes?

Yes! Ask to adjust: “Add a step to…” or “Remove the X part” work perfectly.

Should I always use —dangerously-skip-permissions?

Use when you trust Claude and speed matters. Avoid for production code or when learning new projects.

Do think keywords work in all modes?

Yes! They adjust analysis depth in edit, auto-accept, and plan modes.

🎉 What’s Next?

Level 1 Complete

Congratulations! You’ve mastered all of Level 1: Foundation.

Skills acquired:

  • ✅ File operations (read, write, edit, reference)
  • ✅ Real-time visualization with Obsidian
  • ✅ PM document processing
  • ✅ Parallel agents for cloning yourself
  • ✅ Specialized sub-agent team
  • ✅ Permanent project memory (CLAUDE.md)
  • ✅ Workflow mode selection
  • ✅ Analysis depth control
  • ✅ Friction-free flow state work

Time savings: 10-15 hours per week

Level 2: Real PM Workflows

Ready to save 20+ hours per week? Level 2 teaches daily PM workflows.

Module 2.1: Write a PRD

  • Transform rough ideas into complete PRDs
  • Research agents and Socratic refinement

Module 2.2: Analyze Product Data & Build Dashboards

  • Handle large datasets (CSV, surveys)
  • Multi-source research synthesis

Module 2.3: Build Product Strategy & Roadmap

  • Competitive research at scale
  • Strategic synthesis and roadmap creation

Start Level 2:

/start-2-1

庆祝这个成就！ 你已经为 Claude Code 精通构建了完整的基础。🎓

1.6: 项目记忆2.1: 撰写 PRD