@llms-sdk/prompt

Convention-based prompt template system with multi-step workflows, parking lot processing, and full SDK integration.

Transform your AI workflows from one-off scripts to maintainable, discoverable prompt templates that scale from rapid experimentation to production deployment.

✨ Key Features

🎯 Convention-Based Discovery - Folder name = command name 🔄 Parking Lot Workflow - Chronological iteration through prompts 📊 Multi-Step Sequences - Variable carryforward between phases 🧠 Complexity Analysis - AI-powered breakdown suggestions 🛠️ Development Mode - Full @llms-sdk/* integration 📦 Production Ready - Stable CLI for deployment

🚀 Quick Start

Installation

npm install @llms-sdk/prompt

Your First Prompt

# List available prompts
npx @llms-sdk/prompt list

# Execute a prompt interactively
npx @llms-sdk/prompt quick-security-check

# Use variables directly
npx @llms-sdk/prompt security-audit --variables '{"target_system":"API","audit_scope":"full"}'

Create Your Own

# Single file prompt
echo "Analyze {target} for {criteria}" > prompts/my-analysis.md
npx @llms-sdk/prompt my-analysis

# Multi-step sequence
mkdir -p prompts/my-workflow
echo "Phase 1: {input}" > prompts/my-workflow/01-discovery.md
echo "Phase 2: {step_1_output}" > prompts/my-workflow/02-analysis.md
npx @llms-sdk/prompt my-workflow

🎭 Three Discovery Patterns

graph LR
    A[prompts/] --> B[📄 Single Files]
    A --> C[📁 Folder Structure]
    A --> D[🔢 Multi-Step Sequence]

    B --> B1[quick-check.md]
    C --> C1[build-analysis/01-prompt.md]
    C --> C2[build-analysis/02-runner.ts]
    D --> D1[security-audit/01-recon.md]
    D --> D2[security-audit/02-scan.md]
    D --> D3[security-audit/03-report.md]

    style B fill:#e1f5fe
    style C fill:#f3e5f5
    style D fill:#fff3e0

📚 Complete Documentation

📖 Getting Started Guide

Step-by-step tutorial from installation to advanced workflows

📋 User Guide

Comprehensive feature documentation with examples

🏗️ Architecture Guide

System design, discovery patterns, and execution flows

🎯 Examples Collection

Real-world use cases and patterns

📚 API Reference

Complete CLI commands and options

🔄 Workflow Examples

Parking Lot Processing

# Process all single .md files chronologically (oldest first)
npx @llms-sdk/prompt parking-lot

# Filter by keywords and age
npx @llms-sdk/prompt parking-lot --include="security,audit" --max-age=30

Multi-Step Sequences

# Execute 5-phase security audit with variable carryforward
npx @llms-sdk/prompt security-audit
# → 01-reconnaissance → 02-vulnerability-scan → 03-exploit-analysis → 04-mitigation → 05-report

Development Mode

# Full SDK integration for rapid iteration
tsx scripts/prompt-dev.ts build-failure-analysis

# Template-only processing
tsx scripts/prompt-dev.ts security-audit --template-only

🎨 Template Variables

Templates use {variable} syntax with automatic discovery:

Analyze the **{target_system}** for security vulnerabilities.

Scope: {audit_scope}
Timeline: {completion_date}
Previous findings: {step_1_output}

The system automatically:

• ✅ Extracts all {variables} from templates
• ✅ Prompts interactively for missing values
• ✅ Validates all variables are provided
• ✅ Carries forward outputs between sequence steps

🛠️ Development Integration

Create 02-local-runner.ts for full SDK access:

import { createClient } from "@llms-sdk/core";
import { createBashTool } from "@llms-sdk/toolkit";

export async function run(variables: Record<string, string>, renderedPrompt: string) {
	const client = createClient("anthropic");

	const result = await client.ask({
		messages: [{ role: "user", content: renderedPrompt }],
		tools: [createBashTool()],
	});

	// Execute tools, analyze results, iterate rapidly
	console.log("Result:", result.content);
}

🎯 Use Cases

• 🔍 Code Analysis - Build failures, security audits, performance reviews
• 📋 Documentation - API docs, architecture reviews, compliance reports
• 🚀 DevOps Workflows - Deployment checks, monitoring setup, incident response
• 🧪 Research - Market analysis, competitive research, technical evaluations
• 📊 Data Processing - Log analysis, metrics review, trend identification

🤝 Contributing

1. Follow naming conventions - Folder name = command name
2. Test both CLIs - Stable (npx @llms-sdk/prompt) and development (tsx scripts/prompt-dev.ts)
3. Add comprehensive metadata - Use meta.json for documentation
4. Include development runners - Add 02-local-runner.ts for SDK integration
5. Document variables clearly - Explain expected values and formats

📄 License

MIT License - see LICENSE for details.

🔗 Related Packages

• @llms-sdk/core - Multi-provider LLM client
• @llms-sdk/toolkit - Tool collection and MCP integration
• @llms-sdk/terminal - Terminal UI framework

Testing prp_runner.py and prp_runner.ts

To verify that both the Python and TypeScript PRP runners behave identically, run the following from the root of this package:

bash project/PRPs/scripts/test_prp_runner.sh

Prerequisites:

• Python 3
• Node.js with npx and ts-node
• jq (for JSON comparison)

What it does:

• Creates a test PRP file and a dummy model
• Runs both runners and compares their outputs
• Prints ✅ Outputs match! if they are identical (ignoring session_id)

All test artifacts are cleaned up automatically.

Ready to transform your AI workflows? Start with the Getting Started Guide! 🚀