Claude Code Prompt Cheat Sheet
For deep work, full-codebase analysis, and smart agentic workflows — powered by natural language as the new programming language, with AI agents handling tasks that were once done by humans.
Deep Thinking / Planning
- ultrathink: Plan the architecture for a modular AI tagging system.- think hard: Propose a secure architecture for a multi-tenant Claude plugin platform.- plan: Break this task into phases and estimate time per phase. You have 3 days and solo bandwidth.Audit / Diagnose / Critique
- diagnose: Traverse the codebase and list any patterns that could lead to bugs, confusion, or maintenance issues.- audit: Review my auth.js for potential security issues or logic errors.- You are an external, professional auditing firm specializing in fintech. Perform a complete end-to-end audit of [project]. Produce a formal report (exec summary, scope, methodology, detailed findings, recommendations, conclusion). ultrathink.Understand / Explain / Document
- explain: Walk me through how this app is structured, as if I’m onboarding as a new dev tomorrow.- translate: Explain this entire codebase as if I’m a product manager who understands logic but doesn’t code.- build the narrative: Based on file structure, naming, and content, reconstruct the development history of this project.- doc: Create a missing manual for this codebase with project summary, key modules, config, risks, and extension points.Index / Summarize Files
- ultrathink: Read through the entire project. For each file, summarize what it does in 1–3 sentences. Group by folder.- Tree the file structure and annotate each path with its purpose in 1 sentence.- Search for ‘generateMetadata’ across the entire project. Show where it’s defined, used, and if logic varies.Refactor / Optimize / Rewrite
- refactor: Simplify this transcription.py file to be more modular and testable.- rewrite: Improve the structure and clarity of this README or CLAUDE.md per our style guide.- optimize: Improve performance, clarity, or architecture of this SQL query.Compare / Evaluate Options
- compare: Evaluate three ways to implement feature X—feature flags, component forks, core rewrite—cover pros, cons, maintainability.- spec: Create a detailed CLAUDE.md template for an AI blog editor with live Markdown preview.Role / Mindset
- job: You’re a senior developer reviewing this code for reuse. Suggest improvements and explain why.- You're a staff-level React engineer. Critique component separation and file structure.- You’re a DevOps engineer reviewing Dockerfile and deploy.sh. Suggest improvements based on best practices.Interactive / Debug / Self-Review
- live self-debug mode: Here’s the code you gave me earlier — it’s throwing an error. Act as though you're debugging your own code. Think aloud, step-by-step.- You are a project lead who’s been away. When I return, summarize what we’ve covered and what’s next.- I'm stepping away. When I come back, give me a summary of what’s done, what’s in progress, and any blockers.Simulated Interviews & Q&A
- simulate: You’re a senior dev interviewing me on this codebase. Ask 5 questions testing my understanding of architecture and decisions.Context / Analysis Mode
- shadow: As I paste code edits, watch silently and only speak if significant bugs or patterns emerge.- simulate alternative flow: Simulate both success and failure paths of the login flow, including edge cases.- teach: Explain this decorator pattern with vivid analogies and practical examples.- summarize: Compress this 2,000‑line log file into a categorized bug report.Use these to supercharge Claude into a strategic, structured thinking machine. No slash commands needed — just mindset, role, and intent.