What Is Claude Code?
Anthropic's agentic AI tool that lives in your terminal. It reads your entire codebase, edits files, runs commands, and manages git — all through natural language. Despite the name, it's not just for developers. PMs use it to prototype from PRDs, explore codebases safely, automate workflows, and build internal tools — without writing a single line of code.
Three Surfaces — Same Brain
Claude has three interfaces. Each builds on the last. If you've used Cowork, you're already 70% to Code.
- Conversations & Q&A
- Brainstorming & ideation
- No files, no workflow
- Free tier available
- Extended thinking
- Files & workflows
- Real .docx/.pptx/.xlsx output
- Parallel sub-agents
- Web connectors (Gmail, Slack)
- Sandboxed VM — safe
- Full terminal & git access
- Direct filesystem access
- Plan Mode & subagents
- MCP servers (150+)
- Custom Skills & plugins
Getting Started — 6 Steps
CLAUDE.md & Skills — Your Power Tools
A file at your project root that persists across sessions. Claude reads it on every start. Create with claude init or manually.
- Stack & architecture — frameworks, languages, structure
- Naming conventions — file names, variables, components
- Key file paths — where important code lives
- Testing standards — what to test, how to run
- Common workflows — deploy steps, PR process
- Anti-patterns — what NOT to do
Skills (SKILL.md)
name: write-prd
description: Generate a structured PRD
from a one-line brief. Trigger
when user says "write a PRD".
---
# Role
You are a senior PM writing a PRD.
# Rules
- Use the template in templates/prd.md
- Include user stories, success metrics
- Ask clarifying questions first
Description is the trigger. Claude reads it to decide when to invoke. 200 char max. Write it once → use it hundreds of times.
Essential Commands
| Command | What It Does |
|---|---|
| /plan | Read-only analysis mode — Claude plans, you approve |
| /clear | Wipe conversation history for a fresh start |
| /compact | Compress context without losing essential info |
| /model | Switch between Sonnet (fast) / Opus (deep) / Haiku |
| /effort | Adjust thinking depth — from quick to thorough |
| Command | What It Does |
|---|---|
| /memory | Edit your CLAUDE.md memory files |
| /init | Create a CLAUDE.md template for your project |
| /review | Review a pull request with structured feedback |
| /config | Open settings and preferences |
| /loop | Run a task on a recurring interval (cron-style) |
Context Management — The #1 Skill
Claude's context window holds your entire conversation + every file read + every command output. It fills up fast. The best users obsessively manage it.
/compact to compress the conversation. Add focus: /compact retain error handling patterns/clear. Clean context = better answers. Don't drag unrelated context across tasks.SPEC.md, then start a new session to execute. Clean context focused entirely on implementation.PM Use Cases
MCP Servers & Pro Tips
- GitHub PR workflows, issue management, code review BUILT-IN
- Linear / Jira Query tickets, update status, link to PRDs
- Slack Summarize threads, draft messages, search history
- Notion Reference docs, update pages, query databases
- Figma Inspect designs, design-to-code, extract tokens
- Google Drive Read/write docs and sheets directly
- Playwright Browser automation, testing, screenshots
- 1Plan Mode first. Always. Claude proposes, you approve. No surprises.
- 2One Skill per task. Keep Skills lean and single-purpose. 10 min setup = 40 min saved per use.
- 3Auto Memory works. Claude saves patterns per workspace in MEMORY.md. Use /memory to review.
- 4Sonnet for 80%, Opus for 20%. Sonnet is fast and cheap. Switch to Opus for architecture and refactoring.
- 5Remote control. Start claude in terminal, continue from phone via claude.ai.
- 6Combine Cowork + Code. Use /schedule for daily cron, /loop for in-session monitoring.