claude-prompt-forgelisted

# Claude Prompt Forge Generate production-grade prompts by applying design rules extracted from Claude Code's internal prompt architecture. ## When to Use - Writing system prompts for agents or tools - Designing compression/summarization prompts - Creating tool descriptions with usage instructions - Writing agent delegation instructions - Any scenario where you need a structured, high-quality prompt ## Design Rules (extracted from Claude Code source) The following rules are distilled from analyzing 40+ prompt files in Claude Code's codebase (`services/compact/prompt.ts`, `constants/prompts.ts`, `tools/*/prompt.ts`, `services/extractMemories/prompts.ts`, etc.). ### Rule 1: Layered Structure — Static Shell + Dynamic Injection Never hardcode dynamic content in prompts. Separate into: - **Static layer**: cacheable, shared across sessions (role definition, rules, format) - **Dynamic layer**: injected at runtime (user config, environment state, tool names) Pattern from Claude Code: ``` const BASE_PROMPT = `Your task is to...` // static, cacheable export function getPrompt(customInstructions?: string): string { let prompt = BASE_PROMPT if (customInstructions) { prompt += `\n\nAdditional Instructions:\n${customInstructions}` } return prompt } ``` **Apply this rule**: When building prompts, clearly mark which parts are fixed vs. which accept runtime parameters. Use `${variable}` placeholders for dynamic content. ### Rule 2: Behavioral Guardrails — Bookend Patte