shipdeck/tasks/001_anthropic_api_integration.md

Ship MVPs in 48 hours. Fix bugs in 30 seconds. The command deck for developers who ship.

# Task 001: Anthropic API Integration

## 1. Task Overview

### Task Title
**Title:** Integrate Anthropic SDK for AI-powered code execution

### Goal Statement
**Goal:** Enable Shipdeck Ultimate to execute actual AI-powered code generation through the Anthropic API, transforming it from a template generator to a real MVP builder. This is the foundation that enables our 48-hour guarantee.

---

## 2. Strategic Analysis

### Problem Context
Currently, Shipdeck generates templates but doesn't execute AI tasks. We need to integrate the Anthropic SDK to enable actual code generation and execution.

### Recommendation
Implement direct Anthropic API integration with proper authentication, token management, and streaming responses.

---

## 3. Technical Requirements

### Functional Requirements
- User can provide Anthropic API key via environment variable or config
- System executes AI agents with real Claude API calls
- Token usage tracking and cost estimation
- Streaming responses for real-time progress
- Error handling for API limits and failures

### Non-Functional Requirements
- **Performance:** Sub-second response initiation
- **Security:** Secure API key storage, never exposed in logs
- **Reliability:** Retry logic for transient failures
- **Cost Management:** Token usage tracking and limits

---

## 4. Implementation Plan

### Phase 1: SDK Integration
- [ ] Install @anthropic-ai/sdk package
- [ ] Create lib/ultimate/anthropic-client.js module
- [ ] Implement authentication with API key management
- [ ] Add configuration for model selection (Claude 3.5 Sonnet)

### Phase 2: Agent Execution Layer
- [ ] Create lib/ultimate/agent-executor.js
- [ ] Implement prompt construction from templates
- [ ] Add streaming response handler
- [ ] Implement token counting and cost calculation

### Phase 3: Configuration Management
- [ ] Add API key configuration commands
- [ ] Create config validation
- [ ] Implement usage limits and quotas
- [ ] Add cost estimation features

### Phase 4: Testing
- [ ] Test with simple "Hello World" generation
- [ ] Verify token counting accuracy
- [ ] Test error handling scenarios
- [ ] Validate streaming responses

---

## 5. Success Criteria
- [ ] API key can be configured and validated
- [ ] Simple prompt generates actual code via Claude API
- [ ] Token usage is tracked and displayed
- [ ] Errors are handled gracefully with clear messages
- [ ] Cost estimates are accurate within 5%

---

## 6. Dependencies
- @anthropic-ai/sdk npm package
- Valid Anthropic API key for testing
- Claude 3.5 Sonnet API access

---

## 7. Estimated Effort
**Priority:** P0 (Critical - Week 1)
**Complexity:** Medium
**Duration:** 2-3 days