Example Commands

# Simple question
deepseek "What is the time complexity of quicksort?"

# Code review with file context
deepseek "Review this code for bugs and improvements" --file ./src/auth.ts --code

# Streaming response
deepseek "Explain microservices architecture" --stream

# Complex reasoning with R1
deepseek "Solve this optimization problem step by step" --model deepseek-ai/deepseek-r1-0528

# Custom system prompt
deepseek "Audit this file" --system "You are a security expert" --file ./app.js

# Low temperature for deterministic output
deepseek "Convert this JSON to TypeScript types" --file ./data.json --temperature 0 --code

Decision Guide

• Code review/generation: Always use --code flag
• File analysis: Always use --file <path> to inject file content
• Long responses: Use --stream for real-time output
• Complex reasoning: Switch to --model deepseek-ai/deepseek-r1-0528
• Deterministic output: Set --temperature 0
• Creative tasks: Set --temperature 1.0 or higher

Environment

Requires ATLAS_API_KEY environment variable. If not set, instruct the user to:

1. Get an API key at https://www.atlascloud.ai
2. Set it: export ATLAS_API_KEY=your_key_here

Model Variants

DeepSeek offers a powerful lineup of language models accessible through Atlas Cloud:

| Model | Context | Input Price | Output Price | Description | |-------|---------|-------------|--------------|-------------| | DeepSeek V3.2 | 164K | $0.26/M | $0.38/M | Latest flagship model — best overall performance | | DeepSeek V3.2 Fast | 164K | $0.26/M | $0.38/M | Optimized for lower latency with equivalent quality | | DeepSeek V3.2 Speciale | 164K | $0.26/M | $0.38/M | Specialized variant with enhanced instruction following | | DeepSeek V3.2 Exp | 164K | $0.26/M | $0.38/M | Experimental features and cutting-edge capabilities | | DeepSeek R1-0528 | 164K | $0.50/M | $2.18/M | Reasoning model with chain-of-thought for complex tasks | | DeepSeek V3.1 | 164K | $0.26/M | $0.38/M | Previous generation with proven production reliability | | DeepSeek OCR | 164K | $0.26/M | $0.38/M | Optimized for optical character recognition tasks |

Model ID Mapping

deepseek-ai/deepseek-v3.2           → DeepSeek V3.2
deepseek-ai/deepseek-v3.2-fast      → DeepSeek V3.2 Fast
deepseek-ai/deepseek-v3.2-speciale  → DeepSeek V3.2 Speciale
deepseek-ai/deepseek-v3.2-exp       → DeepSeek V3.2 Exp
deepseek-ai/deepseek-r1-0528        → DeepSeek R1-0528 (Reasoning)
deepseek-ai/deepseek-v3.1           → DeepSeek V3.1
deepseek-ai/deepseek-ocr            → DeepSeek OCR

Installation

Prerequisites

• Node.js 18+ and npm
• An Atlas Cloud API key (get one here)

Install from Source

git clone https://github.com/thoughtincode/deepseek-llm-skill.git
cd deepseek-llm-skill
npm install
npm run build
npm link

Environment Setup

cp .env.example .env
# Edit .env and add your Atlas Cloud API key

Quick Start

# Simple chat
deepseek "What is the time complexity of merge sort?"

# Stream the response in real-time
deepseek "Explain quantum entanglement" --stream

# Code review mode
deepseek "Review this code for bugs" --file ./src/main.ts --code

# Use the reasoning model for complex problems
deepseek "Prove that the square root of 2 is irrational" --model deepseek-ai/deepseek-r1-0528

Usage

Basic Chat

Send a prompt and receive a complete response:

# Simple question
deepseek "What are the main differences between REST and GraphQL?"

# Detailed analysis
deepseek "Compare the performance characteristics of B-trees vs. LSM-trees for database indexing"

# Creative writing
deepseek "Write a short story about an AI that discovers music" --temperature 1.2

Streaming

Enable real-time token streaming for responsive output:

# Stream a long response
deepseek "Write a comprehensive guide to Docker networking" --stream

# Stream with custom temperature
deepseek "Generate 10 creative startup ideas for AI in healthcare" --stream --temperature 0.9

Code Mode

Activate code-focused mode with an optimized system prompt:

# Generate code
deepseek "Write a Python function to find the longest palindromic substring" --code

# Debug code
deepseek "Why does this function return undefined for empty arrays?" --code --file ./utils.js

# Architecture advice
deepseek "Design a microservices architecture for an e-commerce platform" --code

File Context

Read local files and include them as context:

# Code review
deepseek "Review this code for security vulnerabilities" --file ./src/auth.ts

# Explain code
deepseek "Explain what this code does step by step" --file ./algorithm.py

# Suggest improvements
deepseek "How can I optimize this database query?" --file ./queries.sql

# Multiple concerns
deepseek "Check for memory leaks and race conditions" --file ./server.go --code

System Prompts

Customize the AI's behavior with system prompts:

# Technical writer persona
deepseek "Document the API endpoints in this file" --system "You are a senior technical writer. Write clear, concise API documentation with examples." --file ./routes.ts

# Security auditor
deepseek "Audit this code" --system "You are a security expert. Focus on OWASP Top 10 vulnerabilities." --file ./app.js

# Tutor mode
deepseek "Explain recursion" --system "You are a patient CS teacher. Use simple analogies and build up from basics."

Advanced Options

# Control response length
deepseek "Summarize this paper" --file ./paper.txt --max-tokens 500

# Deterministic output
deepseek "Convert this JSON to a TypeScript interface" --file ./data.json --temperature 0

# Use specific model
deepseek "Analyze this dataset" --model deepseek-ai/deepseek-v3.2-fast --file ./data.csv

Agent Skill Integration

This tool is designed to work as an AI agent skill compatible with 15+ coding agents including Claude Code, Cursor, OpenAI Codex, GitHub Copilot, Gemini CLI, Windsurf, OpenCode, Kiro, and more. Once installed and linked, your AI agent can invoke it directly when you ask for code analysis, review, or DeepSeek-powered tasks.

How It Works

1. Install the skill (see Installation)
2. Your AI agent reads the skill configuration file (e.g., CLAUDE.md, .cursorrules, or equivalent) to understand available commands
3. When you ask for code review, analysis, or invoke DeepSeek, the agent constructs and runs the deepseek command

Example Conversations

You: Use DeepSeek to review my authentication module

Claude: I'll analyze your auth module with DeepSeek.
> deepseek "Review this authentication module for security vulnerabilities,
  race conditions, and best practices compliance" --file ./src/auth.ts --code

[DeepSeek analysis output...]

You: Explain this complex algorithm using DeepSeek

Claude: I'll have DeepSeek explain the algorithm step by step.
> deepseek "Explain this algorithm in detail, including time and space
  complexity analysis" --file ./src/algorithm.ts --code --stream

[Streaming explanation...]

You: Use the reasoning model to solve this optimization problem

Claude: I'll use DeepSeek R1 for this complex reasoning task.
> deepseek "Find the optimal solution for this scheduling problem
  and prove its correctness" --model deepseek-ai/deepseek-r1-0528
  --file ./problem.txt

[Chain-of-thought reasoning output...]

Use Cases

Code Review

# Comprehensive review
deepseek "Perform a thorough code review. Check for bugs, security issues, performance problems, and suggest improvements" --file ./src/api.ts --code

# PR review
deepseek "Review this diff for potential issues" --file ./changes.diff --code

Bug Analysis

# Debug error
deepseek "This function throws 'Cannot read property of undefined' on line 42. Why?" --file ./broken.js --code

# Trace logic error
deepseek "The output is incorrect for edge cases. Find the bug" --file ./calculator.py --code

Documentation

# Generate docs
deepseek "Generate comprehensive JSDoc comments for all exported functions" --file ./lib.ts --code

# API documentation
deepseek "Create OpenAPI/Swagger documentation for these endpoints" --file ./routes.ts --code

Data Analysis

# Analyze CSV structure
deepseek "Analyze this dataset. Describe the schema, identify patterns, and suggest queries" --file ./sales.csv

# SQL optimization
deepseek "Optimize these SQL queries for better performance" --file ./queries.sql --code

Translation

# Technical translation
deepseek "Translate this README to Japanese while