Cheat Sheet
Master Claude Code workflows, shortcuts, and best practices for effective AI-assisted development. Learn how to leverage planning modes, dev docs, subagents, and slash commands to maximize productivity.
mindset
Treat Claude like a smart junior dev with amnesia; your job is planning, context, and correction.
planning
Start new features in Plan Mode to brainstorm implementation with Claude before writing code. Iterate on the solution until you are happy, then switch to auto mode.
- 1PressShiftTabTabto enter Plan Mode
- 2Brainstorm implementation with Claude
- 3Use@to reference files for context
- 4Iterate until you are satisfied with the solution
- 5PressShiftTabTabagain to enter auto edit mode
- 6Claude implements with minimal intervention
dev docs
Keep these three markdown files updated so you never lose the plot after compaction or long sessions.
- 1Ask Claude to create the three files for your task
- 2Update them before compacting or ending your session
- 3Read them when resuming work
CLAUDE.md
Keep project-specific rules and run instructions in this file. Keep it short with no generic coding standards.
- 1Run
/initto generate the initial file - 2Add project-specific commands and quirks
- 3Keep it under 200 lines
skills
Modular capabilities that Claude automatically uses when relevant. Unlike slash commands, Claude decides when to invoke them based on context.
.claude/skills/[name]/- 1Create folder with SKILL.md file inside
- 2Add name and description in YAML frontmatter
- 3Write instructions for what the skill does
- 4Claude will use it automatically when appropriate
subagents
Specialized AI assistants with their own separate context window. Each has custom instructions and specific tools for focused tasks.
- 1Run
/agentsto open agent management - 2Click “Create New Agent”
- 3Define purpose, allowed tools, and system prompt
- 4Save to .claude/agents/[name].md
- 5Invoke with “Ask the [name] agent to...”
slash commands
Built-in and custom commands you explicitly call. Create your own reusable workflows or use the many built-in commands.
/initCreate CLAUDE.md file for your project/clearClear conversation history/compactCompress conversation with focus instructions/memoryEdit CLAUDE.md memory file/agentsManage custom AI subagents/reviewRequest code review/helpGet usage help/todosList current todo items/contextView context usage as colored grid/rewindRewind conversation and code/statuslineConfigure status line UI/configOpen settings interface/modelSelect or change AI model/add-dirAdd additional working directories/mcpManage MCP server connections/bashesList and manage background tasks/permissionsView or update permissions/costDisplay token usage analytics/usageShow plan limits and rate status/statusDisplay version and connectivity info/exportExport conversation to file or clipboard/hooksManage hook configurations for tools/sandboxEnable isolated bash execution/output-styleSet output presentation format/terminal-setupInstall keyboard shortcuts/vimEnter vim mode for insert/command modes/doctorCheck installation health status/bugReport bugs to Anthropic/privacy-settingsManage privacy preferences/pr_commentsView pull request feedback/loginSwitch Anthropic accounts/logoutSign out from account/exitExit the sessionprompting
Be specific about what to do, ask for options with pros and cons, avoid leading questions, and reset context when starting fresh.
daily moves
tl;dr
Plan first in Plan Mode, document progress in dev docs, keep project-specific rules in CLAUDE.md, use skills for automatic helpers, subagents for specialized tasks, and slash commands for workflows.
Putting It All Together: A Complete Workflow
Here's a step-by-step guide on how to approach any project using Claude Code, combining all the concepts above into a proven workflow.
Getting Started
Existing Project:
- Run
/initto generate your CLAUDE.md - Customize it with project-specific commands, architecture, and quirks
- Keep it under 200 lines—no generic boilerplate
New Project:
- Press Shift + Tab twice to enter Plan Mode
- Explain what the app should do: "I want to build a task manager app with user auth and real-time updates"
- Discuss tech stack, architecture, and initial features with Claude
- Once aligned, ask Claude to create CLAUDE.md with project setup instructions
Set Up Documentation
Create a docs/ folder with comprehensive project documentation:
- branding.md - Brand colors, fonts, logo usage, voice/tone
- features.md - Complete feature list and requirements
- tech-stack.md - Technologies, frameworks, and libraries used
- pricing.md - Pricing tiers, features per plan (if applicable)
- architecture.md - System design, data models, API structure
Ask Claude to create these during the planning phase. Reference them when needed and update them as the project evolves.
Implement One Feature at a Time
Focus on completing one feature fully before moving to the next:
- Enter Plan Mode: Shift + Tab twice
- Tell Claude: "I want to add user authentication. Stay in plan mode."
- Review Claude's proposed approach, adjust scope if needed
- Switch to auto mode and implement the feature
- Test thoroughly, iterate on feedback
- If Claude goes off track, hit Esc Esc to rewind
Good Habits
Commit Regularly:
After completing each feature or significant change:
git add . git commit -m "Add user authentication with email/password" git push origin dev- Use separate branches for dev and prod
- Review staged changes with
git diffbefore committing - Write descriptive commit messages that explain the "why"
Compact or Clear Regularly:
- Run
/compactwhen conversations get long to free up context - Run
/clearto start fresh (update docs first!) - Use
/contextto monitor context usage
Update Documentation:
- Reference
docs/files when implementing features - Update them when decisions change or new features are added
- Keep CLAUDE.md current with any project quirks you discover
Reflect and Improve:
- After shipping features: What worked well? What didn't?
- Update your workflow and documentation based on learnings
- Share insights with your team
Use Subagents for Specialized Tasks
Set up and use specialized agents for focused work:
- Run
/agentsto create custom subagents early in the project - Create a code-reviewer agent to audit changes before merging
- Set up a debugging agent to investigate test failures
- Use a performance-analysis agent for optimization work
- Create team-specific agents for your workflows
Ship Your Project
Ready to deploy? Follow these final steps:
- Run all tests to ensure everything works:
npm test - Create a PR and ask Claude to draft a detailed summary
- Have the code-reviewer agent audit your changes
- Merge to prod branch and deploy
- Update CLAUDE.md with any deployment quirks or production gotchas
The Pattern That Works
This workflow isn't theoretical—it's battle-tested on real projects. The key insight: treat Claude like a brilliant junior developer. You provide direction, architecture, and quality control. Claude handles execution, explores the codebase, and implements with speed.
Master this workflow and you'll 10x your development velocity while maintaining high code quality. The tools are simple: Plan Mode, dev docs, CLAUDE.md, subagents, and slash commands. The results are extraordinary.