DeFi & Crypto Trading — Uniswap

Engineer from Anthropic, Thariq Shihipar, shared a presentation that has been dubbed the **best alternative to $300 courses** for those looking to become developers using Claude. We've gathered life hacks for working with Claude Code in a convenient format. But first, a tip from us: don't forget to take advantage of the free [Anthropic prompt library](#). **⚙️ Setup and First Steps** - To install Claude Code, you only need Node.js. - Start your journey with AI by asking questions about the codebase (Codebase Q&A) rather than editing code. This will teach you how to formulate prompts correctly and understand the model's limitations. - **Use voice input** if you're working on macOS. Enable the dictation feature in Accessibility settings (Accessibility -> Dictation). Double-click the dictation key and speak your prompts – it's faster than typing them out. - **Initial terminal setup**: right after installation, run the terminal setup to enable the Shift + Enter combination for line breaks (this will eliminate the need to use backslashes). - **Authorization**: set up automatic approval for frequently used tools so that the AI doesn't ask for confirmation for every minor action. **💡 Writing Code and Planning** - Before assigning complex coding tasks to the AI, always **ask it to create a plan**. This will prevent situations where the AI wastes context writing thousands of lines of inappropriate code. - **Iteration through feedback**: The AI works best when it has a way to check its work. Provide it with testing tools: run unit tests or integrate with Puppeteer/iOS simulator for creating interface screenshots. This way, Claude can self-correct in 2-3 iterations. - **Automating Git routines**: use the simple command "commit push." Claude Code will analyze the project's commit history, understand the required formatting style, create a branch, commit changes, and create a Pull Request on GitHub. - **Analyzing Git history**: you can ask the AI why this function has so many arguments or why this variable was added. Claude can independently study Git logs, find authors of changes, related commits, and provide a brief summary. **📎 Managing Context and Configuration File** - **Create a file** called claude.md in the root directory of the project. Write down architectural rules, coding standards, frequently used bash commands, and connected MCP servers. The file will automatically load into context at each startup. - **Local context**: you can create a local claude.md file that doesn't need to be committed to Git – it will only contain your personal project settings. - **Local rules for folders**: claude.md files can be placed in subdirectories. The AI will load them only when it starts working with files inside that specific folder. - **Memory command**: during a dialogue, you can enter the # symbol and write a rule (for example, how to properly use a specific tool), and Claude will automatically save it to the claude.md memory. **🎮 Hotkeys** - **Shift + Tab** – switches the AI to auto-accept edits mode for code. Changes will be applied without your confirmation. - **! (Exclamation mark)** – allows you to run a local bash command. Its output will automatically be included in the context of the next interaction with the AI. - **Escape** – instantly stops the AI from generating or editing code at any moment if you notice an error. This is completely safe and does not break the session. - **Escape (twice)** – allows you to go back through the dialogue history. - **Ctrl + R** – shows the entire context and files that the model sees right now. *Save, use, and forward to colleagues* 🆗 #usefulnotes [Link](#)