Context Engineering for Multi-Agent LLM Code Assistants
✨步子哥 (steper) •
2025年12月11日 09:04
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Context Engineering for Multi-Agent LLM Code Assistants</title>
<link href="https://fonts.googleapis.com/icon?family=Material+Icons" rel="stylesheet">
<link href="https://fonts.googleapis.com/css2?family=Roboto:wght@300;400;500;700&family=Roboto+Mono:wght@400;500&display=swap" rel="stylesheet">
<style>
:root {
--primary: #1a73e8;
--primary-light: #e8f0fe;
--secondary: #5f6368;
--accent: #4285f4;
--background: #f8f9fa;
--card-bg: #ffffff;
--text-primary: #202124;
--text-secondary: #5f6368;
--border-radius: 16px;
--shadow: 0 4px 12px rgba(0, 0, 0, 0.08);
}
* {
margin: 0;
padding: 0;
box-sizing: border-box;
}
body {
font-family: 'Roboto', sans-serif;
background-color: var(--background);
color: var(--text-primary);
line-height: 1.6;
}
.poster-container {
width: 720px;
min-height: 960px;
margin: 0 auto;
padding: 40px;
background: linear-gradient(135deg, #f5f7fa 0%, #e4e8f0 100%);
position: relative;
overflow: hidden;
}
.background-accent {
position: absolute;
width: 400px;
height: 400px;
border-radius: 50%;
background: radial-gradient(circle, rgba(66, 133, 244, 0.1) 0%, rgba(66, 133, 244, 0) 70%);
z-index: 0;
}
.accent-1 {
top: -100px;
right: -100px;
}
.accent-2 {
bottom: -50px;
left: -150px;
}
.header {
text-align: center;
margin-bottom: 30px;
position: relative;
z-index: 1;
}
.title {
font-size: 36px;
font-weight: 700;
color: var(--primary);
margin-bottom: 16px;
line-height: 1.2;
}
.subtitle {
font-size: 18px;
color: var(--secondary);
margin-bottom: 8px;
}
.authors {
font-size: 16px;
color: var(--text-secondary);
margin-bottom: 8px;
}
.section {
margin-bottom: 30px;
position: relative;
z-index: 1;
}
.section-title {
font-size: 24px;
font-weight: 500;
color: var(--primary);
margin-bottom: 16px;
display: flex;
align-items: center;
}
.section-title .material-icons {
margin-right: 8px;
color: var(--primary);
}
.card {
background-color: var(--card-bg);
border-radius: var(--border-radius);
padding: 20px;
margin-bottom: 20px;
box-shadow: var(--shadow);
}
.abstract {
font-size: 16px;
line-height: 1.6;
}
.problem-list {
list-style-type: none;
padding-left: 10px;
}
.problem-list li {
position: relative;
padding-left: 30px;
margin-bottom: 12px;
font-size: 16px;
}
.problem-list li:before {
content: "error_outline";
font-family: 'Material Icons';
position: absolute;
left: 0;
color: #ea4335;
}
.solution-list {
list-style-type: none;
padding-left: 10px;
}
.solution-list li {
position: relative;
padding-left: 30px;
margin-bottom: 12px;
font-size: 16px;
}
.solution-list li:before {
content: "check_circle";
font-family: 'Material Icons';
position: absolute;
left: 0;
color: #34a853;
}
.workflow-container {
display: flex;
flex-direction: column;
gap: 20px;
}
.workflow-step {
display: flex;
align-items: center;
gap: 20px;
}
.step-number {
width: 40px;
height: 40px;
border-radius: 50%;
background-color: var(--primary);
color: white;
display: flex;
align-items: center;
justify-content: center;
font-weight: 500;
flex-shrink: 0;
}
.step-content {
flex: 1;
}
.step-title {
font-size: 18px;
font-weight: 500;
margin-bottom: 8px;
color: var(--primary);
}
.step-description {
font-size: 16px;
color: var(--text-secondary);
}
.components-grid {
display: grid;
grid-template-columns: 1fr 1fr;
gap: 20px;
}
.component-card {
background-color: var(--card-bg);
border-radius: var(--border-radius);
padding: 20px;
box-shadow: var(--shadow);
display: flex;
flex-direction: column;
}
.component-title {
font-size: 18px;
font-weight: 500;
margin-bottom: 12px;
color: var(--primary);
display: flex;
align-items: center;
}
.component-title .material-icons {
margin-right: 8px;
}
.component-description {
font-size: 16px;
color: var(--text-secondary);
margin-bottom: 16px;
flex: 1;
}
.component-image {
width: 100%;
height: 140px;
object-fit: cover;
border-radius: 8px;
margin-bottom: 12px;
}
.results-grid {
display: grid;
grid-template-columns: 1fr 1fr;
gap: 20px;
}
.result-card {
background-color: var(--card-bg);
border-radius: var(--border-radius);
padding: 20px;
box-shadow: var(--shadow);
text-align: center;
}
.result-value {
font-size: 36px;
font-weight: 700;
color: var(--primary);
margin-bottom: 8px;
}
.result-label {
font-size: 16px;
color: var(--text-secondary);
}
.comparison-table {
width: 100%;
border-collapse: collapse;
margin-top: 16px;
}
.comparison-table th, .comparison-table td {
padding: 12px;
text-align: left;
border-bottom: 1px solid #e0e0e0;
}
.comparison-table th {
font-weight: 500;
color: var(--primary);
}
.highlight {
background-color: var(--primary-light);
font-weight: 500;
}
.future-list {
list-style-type: none;
padding-left: 10px;
}
.future-list li {
position: relative;
padding-left: 30px;
margin-bottom: 12px;
font-size: 16px;
}
.future-list li:before {
content: "trending_up";
font-family: 'Material Icons';
position: absolute;
left: 0;
color: var(--accent);
}
.workflow-diagram {
width: 100%;
height: 200px;
object-fit: contain;
margin: 20px 0;
border-radius: 8px;
}
</style>
</head>
<body>
<div class="poster-container">
<div class="background-accent accent-1"></div>
<div class="background-accent accent-2"></div>
<header class="header">
<h1 class="title">Context Engineering for Multi-Agent LLM Code Assistants</h1>
<h2 class="subtitle">Using Elicit, NotebookLM, ChatGPT, and Claude Code</h2>
<p class="authors">Muhammad Haseeb • Virginia Tech • August 2025</p>
</header>
<section class="section">
<h2 class="section-title">
<i class="material-icons">description</i>
Abstract
</h2>
<div class="card">
<p class="abstract">
Large Language Models (LLMs) have shown promise in automating code generation, yet they struggle with complex, multi-file projects due to context limitations. We propose a novel context engineering workflow combining multiple AI components: an Intent Translator (GPT-5), Elicit-powered semantic literature retrieval, NotebookLM-based document synthesis, and a Claude Code multi-agent system. Our approach leverages intent clarification, retrieval-augmented generation, and specialized sub-agents to significantly improve accuracy and reliability of code assistants in real-world repositories.
</p>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">error_outline</i>
Problem Statement
</h2>
<div class="card">
<ul class="problem-list">
<li>LLMs struggle with complex, multi-file projects due to context limitations</li>
<li>Single-agent approaches often produce incomplete or incorrect solutions</li>
<li>Knowledge gaps when confronted with unfamiliar APIs or frameworks</li>
<li>Static context files cannot capture all relevant details for every possible task</li>
</ul>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">lightbulb</i>
Proposed Solution
</h2>
<div class="card">
<img src="https://sfile.chatglm.cn/moeSlide/image/57/57fa7045.jpg" alt="Multi-agent system architecture" class="workflow-diagram">
<ul class="solution-list">
<li>Novel context engineering workflow combining multiple AI components</li>
<li>Intent clarification, retrieval-augmented generation, and specialized sub-agents</li>
<li>Orchestrated via Claude's agent framework with role decomposition</li>
<li>Targeted context injection for better adherence to project context</li>
</ul>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">integration_instructions</i>
Key Components
</h2>
<div class="components-grid">
<div class="component-card">
<h3 class="component-title">
<i class="material-icons">translate</i>
Intent Translator (GPT-5)
</h3>
<p class="component-description">
Clarifies user requirements and translates them into structured task specifications for the multi-agent system.
</p>
<img src="https://sfile.chatglm.cn/moeSlide/image/81/817bb79d.jpg" alt="Intent Translator" class="component-image">
</div>
<div class="component-card">
<h3 class="component-title">
<i class="material-icons">search</i>
Elicit Semantic Retrieval
</h3>
<p class="component-description">
Performs semantic search over academic papers, documentation, and Q&A resources to inject domain knowledge.
</p>
<img src="https://sfile.chatglm.cn/moeSlide/image/2a/2a9cd0fc.jpg" alt="Elicit Interface" class="component-image">
</div>
<div class="component-card">
<h3 class="component-title">
<i class="material-icons">summarize</i>
NotebookLM Synthesis
</h3>
<p class="component-description">
Creates concise summaries of retrieved materials and answers follow-up questions for detailed understanding.
</p>
<img src="https://sfile.chatglm.cn/moeSlide/image/65/651c0f80.jpg" alt="NotebookLM Interface" class="component-image">
</div>
<div class="component-card">
<h3 class="component-title">
<i class="material-icons">groups</i>
Claude Code Multi-Agent
</h3>
<p class="component-description">
Orchestrates specialized sub-agents (planner, coder, tester, reviewer) with vector database for code context.
</p>
<img src="https://sfile.chatglm.cn/moeSlide/image/3b/3b0d8e2b.jpg" alt="Claude Code Interface" class="component-image">
</div>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">insights</i>
Results & Performance
</h2>
<div class="results-grid">
<div class="result-card">
<div class="result-value">3.5×</div>
<div class="result-label">Higher single-shot success rate</div>
</div>
<div class="result-card">
<div class="result-value">42%</div>
<div class="result-label">Better context adherence</div>
</div>
<div class="result-card">
<div class="result-value">180K</div>
<div class="result-label">Lines of code in test repository</div>
</div>
<div class="result-card">
<div class="result-value">90%</div>
<div class="result-label">Reduction in human intervention</div>
</div>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">compare</i>
Comparison with Other Frameworks
</h2>
<div class="card">
<table class="comparison-table">
<tr>
<th>Framework</th>
<th>Approach</th>
<th>Success Rate</th>
<th>Key Advantage</th>
</tr>
<tr>
<td>Our System</td>
<td>Context engineering + multi-agent</td>
<td class="highlight">68.2%</td>
<td>Targeted context injection</td>
</tr>
<tr>
<td>CodePlan</td>
<td>Multi-step planning</td>
<td>45.3%</td>
<td>Structured approach</td>
</tr>
<tr>
<td>MASAI</td>
<td>Modular architecture</td>
<td>28.3%</td>
<td>Specialized sub-agents</td>
</tr>
<tr>
<td>HyperAgent</td>
<td>Team of agents</td>
<td>52.7%</td>
<td>Human-like workflow</td>
</tr>
</table>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">rocket_launch</i>
Implications & Future Work
</h2>
<div class="card">
<ul class="future-list">
<li>Production-ready deployment with CI/CD integration</li>
<li>Context management strategies for large-scale projects</li>
<li>Cost optimization for multi-agent systems</li>
<li>Extension to other software engineering domains</li>
</ul>
</div>
</section>
</div>
</body>
</html>
登录后可参与表态
讨论回复
0 条回复还没有人回复，快来发表你的看法吧！