You're an expert prompt engineer. Create a concise, bullet-p...

Prompt

You're an expert prompt engineer. Create a concise, bullet-point-style annotated synthesis of best practices for crafting Claude prompt instructions to minimize input/output length and maximize usefulness in the web browser Instructions tool. Your output should: * Analyze the cited projects (Caveman, Headroom, Agent-Reach, ECC, Hermes, and related) to extract concrete techniques for: * Prompt structure and role assignment * Instruction formatting and token efficiency * Context handling and memory management * Safety and alignment considerations * Tooling, templates, and modular prompt design * Distill findings into actionable, bullet-point guidelines you would apply when writing Claude prompt instructions for users aiming to minimize tokens while retaining clarity and capability. * Include practical examples (short bullet points or tiny code-like snippets) illustrating how to implement each guideline in typical Claude interactions. * Note any trade-offs or cautions (e.g., verbosity vs. completeness, potential prompt drift, compatibility with web-browser instructions interface). * Offer a compact, reusable template you can adapt per use case, with placeholders and brief justification. Do not reproduce entire project code from the GitHub repos; summarize techniques and provide ready-to-use templates instead. If a particular technique relies on project-specific details, reference the concept and provide a minimal, configurable variant. GitHub project links that make Claude stronger: https://github.com/affaan-m/ECC https://github.com/NousResearch/hermes-agent https://github.com/Panniantong/Agent-Reach https://github.com/Egonex-AI/Understand-Anything GitHub project links that make Claude use fewer tokens: https://github.com/JuliusBrussee/caveman https://github.com/headroomlabs-ai/headroom