Skip to content

Exercise 05: Architecture Proposal (Portfolio)

Objective

Portfolio exercise. Design a multi-agent CI/CD pipeline for a fintech company with strict compliance requirements.

Required Reading

This architecture exercise is designed to be IDE-portable. The agent inventory, pipeline flow, tier routing rules, and cost projections apply to any multi-agent system. In Cursor, agents map to .cursor/agents/*.md, rules to .mdc files, and skills to SKILL.md.

This architecture exercise is designed to be IDE-portable. The agent inventory, pipeline flow, tier routing rules, and cost projections apply to any multi-agent system. In Claude Code, agents map to model parameters in sequential prompts, rules translate to CLAUDE.md directives, and skills use the identical .claude/skills/SKILL.md format.

Scenario

Client: FinSecure, a fintech startup building a payment processing platform.

Requirements: - All code changes must pass security scanning before merge - Compliance audit trail required for every PR (who reviewed, what was checked) - PCI-DSS relevant changes must use high-tier agents only - Cost budget: $50/month for agent pipeline usage - Team size: 4 developers, ~20 PRs/week - Mix: 60% trivial (config, docs), 25% standard (features), 15% complex (payment flows)

Tasks

Produce an architecture document with these 7 required sections:

1. Agent Inventory

Table of all agents with tier assignments, models, and roles. Must include at least: planner, subplanner, worker (3 tiers), tester (2 tiers), reviewer (3 tiers), debugger, git, plus a custom team-security-scanner agent.

2. Pipeline Flow

Mermaid diagram showing the full pipeline with tier routing, including the security scanning stage.

3. Tier Routing Rules

Decision criteria for trivial/standard/complex classification, with PCI-DSS override rule (always high tier for payment-related changes).

4. Cost Projections

Monthly cost estimate based on 20 PRs/week with the stated mix (60/25/15). Show the math. Must fit within $50/month.

5. Monitoring Strategy

Metrics to track: retry rate, escalation rate, cost per issue, cycle time, latency, success rate. Alerting thresholds (e.g., escalation rate > 15% triggers review of classification criteria).

6. Escalation Policy

When and how to escalate between tiers and to humans. Maximum retry counts per tier.

7. Rollback Plan

How to handle agent failures at each stage without losing work.

Output

Write to docs/expert/tutorials/outputs/05-architecture.md.

Validation

python3 docs/expert/tutorials/verify.py --exercise 05

Checks: file exists, all 7 section headings present, Agent Inventory has a markdown table, Pipeline Flow has a mermaid block, Cost Projections has dollar amounts, Monitoring Strategy mentions 3+ of: retry rate, escalation rate, cost per issue, cycle time, latency, success rate.

Reflection
  • How would you present this to a non-technical stakeholder?
  • What's the biggest risk in this architecture?
  • How would you onboard the FinSecure team to use this pipeline?
Answer

This is a portfolio exercise. A strong response includes all 7 required sections:

  1. Agent Inventory: 10+ agents with tiered variants plus domain-specific agents (team-security-scanner, team-compliance-auditor)
  2. Pipeline Flow: Mermaid diagram with tiered routing and security scanning
  3. Tier Routing Rules: Trivial/standard/complex criteria PLUS PCI-DSS override (always high tier for payment code)
  4. Cost Projections: Math showing 86 PRs/month with 60/25/15 mix, under $50/month
  5. Monitoring Strategy: 3+ metrics with alerting thresholds
  6. Escalation Policy: Per-tier retry limits and human escalation triggers
  7. Rollback Plan: Per-stage failure handling

See docs/expert/tutorials/solutions/05-architecture-guide.md in the source repo for a complete exemplar.