27 lines
13 KiB
JavaScript
27 lines
13 KiB
JavaScript
import{c as t,Q as a,j as i,m as s}from"./chunks/framework.BPKcPtvA.js";const u=JSON.parse('{"title":"Module 8: Capstone — Production Multi-Agent System","description":"","frontmatter":{},"headers":[],"relativePath":"modules/m8-capstone.md","filePath":"modules/m8-capstone.md","lastUpdated":1780488246000}'),n={name:"modules/m8-capstone.md"};function r(l,e,o,d,h,p){return a(),i("div",null,[...e[0]||(e[0]=[s(`<h1 id="module-8-capstone-—-production-multi-agent-system" tabindex="-1">Module 8: Capstone — Production Multi-Agent System <a class="header-anchor" href="#module-8-capstone-—-production-multi-agent-system" aria-label="Permalink to "Module 8: Capstone — Production Multi-Agent System""></a></h1><h2 id="overview" tabindex="-1">Overview <a class="header-anchor" href="#overview" aria-label="Permalink to "Overview""></a></h2><p>Build, deploy, and evaluate a complete production-grade multi-agent system. Choose from three project options, or propose your own.</p><hr><h2 id="project-options" tabindex="-1">Project Options <a class="header-anchor" href="#project-options" aria-label="Permalink to "Project Options""></a></h2><h3 id="option-1-brand-monitor-intermediate" tabindex="-1">Option 1: Brand Monitor (Intermediate) <a class="header-anchor" href="#option-1-brand-monitor-intermediate" aria-label="Permalink to "Option 1: Brand Monitor (Intermediate)""></a></h3><p>Build a multi-agent system that monitors how LLMs mention a brand across multiple AI services.</p><p><strong>Agents</strong>:</p><ul><li><strong>Scanner Agent</strong> — queries ChatGPT, Gemini, Grok, Perplexity for brand mentions</li><li><strong>Analyzer Agent</strong> — classifies mentions as positive/negative/neutral</li><li><strong>Reporter Agent</strong> — generates daily summary reports</li></ul><p><strong>Requirements</strong>:</p><ul><li>Each agent has domain-locked permissions</li><li>Each agent maintains a mental model</li><li>Scanner uses cascade routing (Flash for scrape, Sonnet for analysis)</li><li>Daily automated runs via heartbeat</li><li>Cost tracking per scan</li></ul><h3 id="option-2-code-review-pipeline-intermediate-advanced" tabindex="-1">Option 2: Code Review Pipeline (Intermediate-Advanced) <a class="header-anchor" href="#option-2-code-review-pipeline-intermediate-advanced" aria-label="Permalink to "Option 2: Code Review Pipeline (Intermediate-Advanced)""></a></h3><p>A plan → build → review → verify chain with CI/CD integration.</p><p><strong>Agents</strong>:</p><ul><li><strong>Planner</strong> — reads spec, creates implementation plan</li><li><strong>Builder</strong> — writes code per plan</li><li><strong>Reviewer</strong> — reviews code for bugs, style, security</li><li><strong>Verifier</strong> — read-only agent that checks all claims</li></ul><p><strong>Requirements</strong>:</p><ul><li>YAML-defined agent chain pipeline</li><li>Golden dataset for regression testing</li><li>Verifier agent with confidence ladder</li><li>GitHub PR integration</li><li>Cost analysis per review cycle</li></ul><h3 id="option-3-strategic-decision-board-advanced" tabindex="-1">Option 3: Strategic Decision Board (Advanced) <a class="header-anchor" href="#option-3-strategic-decision-board-advanced" aria-label="Permalink to "Option 3: Strategic Decision Board (Advanced)""></a></h3><p>An 8-agent CEO board with adversarial debate, fact verification, and execution planning.</p><p><strong>Agents</strong>:</p><ul><li>CEO (orchestrator)</li><li>6 board members with distinct perspectives</li><li>Verifier (fact-checker)</li><li>Tracker (quarterly review)</li></ul><p><strong>Requirements</strong>:</p><ul><li>Each board member has distinct persona + mental model</li><li>Fact verification (2+ sources per claim)</li><li>Contrarian position tracked at 3x weight</li><li>Structured memo output</li><li>Audio summary (TTS)</li><li>Cost tracking per deliberation</li></ul><hr><h2 id="capstone-phases" tabindex="-1">Capstone Phases <a class="header-anchor" href="#capstone-phases" aria-label="Permalink to "Capstone Phases""></a></h2><h3 id="phase-1-system-design-60-90-min" tabindex="-1">Phase 1: System Design (60-90 min) <a class="header-anchor" href="#phase-1-system-design-60-90-min" aria-label="Permalink to "Phase 1: System Design (60-90 min)""></a></h3><p>Deliverable: One-page architecture document including:</p><ul><li>System diagram (agents + data flow + tool surface)</li><li>Agent roster (names, personas, domain permissions)</li><li>Data model (what state is shared, how)</li><li>Security architecture (what L-level, what hooks)</li><li>Cost estimate per session</li></ul><p><strong>Template</strong>:</p><div class="language-markdown vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang">markdown</span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span style="--shiki-light:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold;"># System Design: [Project Name]</span></span>
|
||
<span class="line"></span>
|
||
<span class="line"><span style="--shiki-light:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold;">## Architecture</span></span>
|
||
<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">[ASCII diagram or mermaid]</span></span>
|
||
<span class="line"></span>
|
||
<span class="line"><span style="--shiki-light:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold;">## Agent Roster</span></span>
|
||
<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">| Agent | Model | Tools | Domain | Memory |</span></span>
|
||
<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">|-------|-------|-------|--------|--------|</span></span>
|
||
<span class="line"></span>
|
||
<span class="line"><span style="--shiki-light:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold;">## Data Flow</span></span>
|
||
<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">[How agents communicate, what state they share]</span></span>
|
||
<span class="line"></span>
|
||
<span class="line"><span style="--shiki-light:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold;">## Security</span></span>
|
||
<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">[L-level, hooks, verifier, damage-control rules]</span></span>
|
||
<span class="line"></span>
|
||
<span class="line"><span style="--shiki-light:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold;">## Cost Estimate</span></span>
|
||
<span class="line"><span style="--shiki-light:#24292E;--shiki-dark:#E1E4E8;">[Per-session estimate with model pricing]</span></span></code></pre></div><h3 id="phase-2-agent-implementation-2-4-hours" tabindex="-1">Phase 2: Agent Implementation (2-4 hours) <a class="header-anchor" href="#phase-2-agent-implementation-2-4-hours" aria-label="Permalink to "Phase 2: Agent Implementation (2-4 hours)""></a></h3><p>Build each agent:</p><ul><li>System prompt with persona + behavioral rules</li><li>Tool definitions with reasoning parameter</li><li>Mental model file (agent-owned expertise)</li><li>Skills for composable behavior</li></ul><h3 id="phase-3-integration-1-2-hours" tabindex="-1">Phase 3: Integration (1-2 hours) <a class="header-anchor" href="#phase-3-integration-1-2-hours" aria-label="Permalink to "Phase 3: Integration (1-2 hours)""></a></h3><p>Wire agents together:</p><ul><li>Chain (if pipeline pattern)</li><li>Team (if dispatcher pattern)</li><li>P2P (if flat pattern)</li><li>Shared context (conversation log)</li></ul><h3 id="phase-4-security-hardening-1-2-hours" tabindex="-1">Phase 4: Security Hardening (1-2 hours) <a class="header-anchor" href="#phase-4-security-hardening-1-2-hours" aria-label="Permalink to "Phase 4: Security Hardening (1-2 hours)""></a></h3><p>Apply security layer:</p><ul><li>Minimum L3 (blacklist hook)</li><li>Target L4 (whitelist hook)</li><li>Optional L5 (no bash)</li><li>Damage-control rules</li><li>Verifier agent (read-only checks)</li></ul><h3 id="phase-5-testing-evaluation-1-2-hours" tabindex="-1">Phase 5: Testing & Evaluation (1-2 hours) <a class="header-anchor" href="#phase-5-testing-evaluation-1-2-hours" aria-label="Permalink to "Phase 5: Testing & Evaluation (1-2 hours)""></a></h3><ul><li>Create golden dataset (10+ test cases)</li><li>Run pass@k evaluation (k=3, k=5)</li><li>Calculate cost per task</li><li>Identify optimization opportunities</li></ul><h3 id="phase-6-deployment-1-2-hours" tabindex="-1">Phase 6: Deployment (1-2 hours) <a class="header-anchor" href="#phase-6-deployment-1-2-hours" aria-label="Permalink to "Phase 6: Deployment (1-2 hours)""></a></h3><ul><li>CI/CD pipeline with golden dataset gate</li><li>Shadow deployment vs baseline</li><li>Observability: every tool call + cost logged</li><li>Monitoring dashboard</li></ul><h3 id="phase-7-review-iterate-1-hour" tabindex="-1">Phase 7: Review & Iterate (1 hour) <a class="header-anchor" href="#phase-7-review-iterate-1-hour" aria-label="Permalink to "Phase 7: Review & Iterate (1 hour)""></a></h3><ul><li>Red-team your own system (try to break it)</li><li>Document lessons learned</li><li>Create retrospective</li></ul><hr><h2 id="deliverables-checklist" tabindex="-1">Deliverables Checklist <a class="header-anchor" href="#deliverables-checklist" aria-label="Permalink to "Deliverables Checklist""></a></h2><div class="language- vp-adaptive-theme"><button title="Copy Code" class="copy"></button><span class="lang"></span><pre class="shiki shiki-themes github-light github-dark vp-code" tabindex="0"><code><span class="line"><span>[ ] Architecture document (one page, with diagram)</span></span>
|
||
<span class="line"><span>[ ] All agent source code with system prompts</span></span>
|
||
<span class="line"><span>[ ] Mental model files for each agent</span></span>
|
||
<span class="line"><span>[ ] Security configuration (damage-control rules)</span></span>
|
||
<span class="line"><span>[ ] Verifier agent (if applicable)</span></span>
|
||
<span class="line"><span>[ ] Golden dataset (10+ test cases)</span></span>
|
||
<span class="line"><span>[ ] Test results (pass@k at k=1, k=3, k=5)</span></span>
|
||
<span class="line"><span>[ ] Cost analysis ($/task, optimization opportunities)</span></span>
|
||
<span class="line"><span>[ ] Security audit (which L-level, what gaps remain)</span></span>
|
||
<span class="line"><span>[ ] Retrospective (max 1 page)</span></span></code></pre></div><hr><h2 id="pass-criteria" tabindex="-1">Pass Criteria <a class="header-anchor" href="#pass-criteria" aria-label="Permalink to "Pass Criteria""></a></h2><table tabindex="0"><thead><tr><th>Criterion</th><th>Minimum</th><th>Target</th></tr></thead><tbody><tr><td>System runs without manual intervention</td><td>✓</td><td>✓</td></tr><tr><td>All agents have domain-locked permissions</td><td>✓</td><td>✓</td></tr><tr><td>Each agent has mental model file</td><td>✓</td><td>✓</td></tr><tr><td>pass@k (k=3) on golden dataset</td><td>>60%</td><td>>80%</td></tr><tr><td>Cost analysis within 2x of optimal</td><td>✓</td><td>✓</td></tr><tr><td>Security audit identifies ≥2 improvements</td><td>✓</td><td>✓</td></tr><tr><td>Observability captures all tool calls</td><td>✓</td><td>✓</td></tr><tr><td>Architecture document submitted</td><td>✓</td><td>✓</td></tr></tbody></table><hr><h2 id="grading-rubric" tabindex="-1">Grading Rubric <a class="header-anchor" href="#grading-rubric" aria-label="Permalink to "Grading Rubric""></a></h2><table tabindex="0"><thead><tr><th>Area</th><th>Weight</th><th>Poor (0)</th><th>Good (1)</th><th>Excellent (2)</th></tr></thead><tbody><tr><td>Architecture</td><td>20%</td><td>No diagram, unclear design</td><td>Diagram present, mostly clear</td><td>Clear diagram, justified choices</td></tr><tr><td>Implementation</td><td>25%</td><td>Agents don't work</td><td>Agents work on happy path</td><td>Agents handle errors gracefully</td></tr><tr><td>Security</td><td>20%</td><td>L1 only</td><td>L3+ with damage-control</td><td>L4+ with verifier</td></tr><tr><td>Testing</td><td>15%</td><td>No eval</td><td>pass@k computed</td><td>pass@k + cost analysis + grind detection</td></tr><tr><td>Documentation</td><td>10%</td><td>Minimal</td><td>Architecture + setup</td><td>Architecture + setup + retrospective</td></tr><tr><td>Cost Optimization</td><td>10%</td><td>Single model</td><td>Cascade routing</td><td>Cascade + verified savings</td></tr></tbody></table>`,54)])])}const g=t(n,[["render",r]]);export{u as __pageData,g as default};
|