vrajpal-jhala

Prompting Patterns for Coding Agents

Interaction patterns I use with AI coding agents (Claude Code, in my case), pulled from real sessions across several projects. Not "best prompts" — just how I keep an agent pointed the right direction instead of running off on its own.

1. Separate "figure it out" from "do it"

Asking a question reads as permission to act unless you say otherwise.

"How should I create a background thread which I can stream whenever I want? don't jump to implementation" "the repo_instructions node doesn't carry any data, should it? if so, what? don't make changes"

vrajpal-jhala

#!/bin/sh
#
# Cursor AI IDE Installer/Updater/Uninstaller
# This script installs, updates, or uninstalls Cursor AI IDE on Linux systems
#

# -----------------------------------------------------------------------------
# Configuration
# -----------------------------------------------------------------------------
APPIMAGE_PATH="/opt/cursor/cursor.AppImage"