-
-
Hacker Challenge - race the hacker to protect the Bedrock Agentcore Service before its exploited
-
AWS Legend Jeff Barr playing Zombie Blaster at re:Invent!
-
My 9yo son learning what an unused AWS Identity is :)
-
Lobby AKA AWS Organization View
-
Regular Gameplay
-
Game Over - Zombies are set free with the API
-
Scattered Spider Level Boss
-
Audit Challenge - Protect the Admin Access in production or receive a significant deficiency
-
Production Outage! Cannot do anything for 5 seconds!
🧟 Sonrai Zombie Blaster
Turn cloud security into an 8-bit adventure—eliminate zombie IAM identities with real API calls
🎯 TL;DR for Judges
| What | Evidence |
|---|---|
| The Game | 8-bit arcade game where zapping zombies triggers real Sonrai API calls to quarantine unused AWS identities |
| Kiro Usage | All 5 features: vibe coding, specs, steering, hooks, MCP |
| Unique Factor | 12-agent Architecture Review Board—Kiro as PM, Architect, QA Lead, Security Lead, and 8 more |
| Results | Built in 11 days • 667 tests • 60 FPS • Zero P0 bugs |
⏱️ 60-Second Demo: Clone → python3 src/main.py → Type ARCADE → Blast zombies → Watch real API calls quarantine cloud identities
💡 The Inspiration
Cloud security and Identity hygiene are overwhelming.
After years of watching cloud and security teams struggle to correct "unused IAM identities", I asked: What if it were a game?
The lobby visualizes your AWS organization—each door is a real AWS account
The unintended benefit; during development and testing, my 9 year old son got interested and began learning what unused identities means and a new QA tester was born. Ultimately my 4 year old twins saw him playing and wanted to join in the fun, so I happily outsourced game testing to my 3 boys. This has been such a fun project for our whole family.
The metaphor clicked instantly:
- 🧟 Zombies = Unused AWS identities (the "dead" accounts that should be eliminated)
- 🔫 Shooting zombies = Real API calls that quarantine actual cloud resources
- 🛡️ Purple shields = Protected identities (Sonrai's Cloud Permissions Firewall)
- 🎮 Retro gaming = Making complex topics approachable and fun
The resurrection works on two levels: We resurrected retro gaming AND we're eliminating "zombie" identities that should be dead.
Zombies are real unused identities—purple shields show protected resources
📊 By The Numbers
┌─────────────────────────────────────────────────────────────┐
│ 🚀 BUILT IN 11 DAYS WITH KIRO │
├─────────────────────────────────────────────────────────────┤
│ 8,000 lines of code │ 667 automated tests │
│ 60 FPS (4× improvement) │ 15+ Sonrai API operations │
│ 15+ steering files │ 6 complete feature specs │
│ 7 agent hooks │ 43 documentation files │
│ 12-agent ARB review │ Zero P0 bugs in production │
└─────────────────────────────────────────────────────────────┘
🤖 How I Used Kiro (All 5 Features)
1. 💬 Vibe Coding: 45 Minutes Instead of 8 Hours
The Challenge: Build arcade mode with dynamic spawning, combo system, and batch quarantine.
The Conversation:
Me: "I need an arcade mode where zombie spawn rates increase over time, with combo multipliers and a results screen"
Kiro: Generated 800 lines of production code with state machine, dynamic spawning algorithm, combo tracking, and results screen
Arcade mode results—built in 45 minutes with Kiro
The Impact:
| Traditional | With Kiro | Savings |
|---|---|---|
| 6-8 hours | 45 minutes | 90% |
Other Vibe Coding Wins:
- JIT Access Quest with auditor patrol AI
- Photo booth with retro pixel filters
- Spatial grid collision (15 FPS → 60 FPS)
2. 📋 Specs: Zero P0 Bugs Through Structure
The Process:
requirements.md → design.md → tasks.md → implementation → tests
↓ ↓ ↓ ↓ ↓
User stories Architecture 15 steps 800 lines 105 tests
6 Complete Specs:
| Spec | What It Does | Tests |
|---|---|---|
| Arcade Mode | 60-second timed challenge with combos | 105 |
| Service Protection Quest | Race hackers to protect services | 25 |
| JIT Access Quest | Protect admin roles from auditors | 25 |
| Photo Booth | Retro selfies with game stats | 20 |
| Account Wall Defense | Tower defense mechanics | 30 |
| Level Progression | Unlock system with difficulty scaling | 15 |
![]() JIT Access Quest—protect admin roles |
![]() Service Protection—race the hacker |
The Magic: Property-based tests caught 3 edge cases before they became bugs.
3. 📜 Steering: From Generic to Domain Expert
15+ Steering Files Defining Agent Roles:
.kiro/steering/
├── product-manager.md # Sprint planning, backlog prioritization
├── architecture-agent.md # Design patterns, performance standards
├── security-agent.md # SAST, secret detection, best practices
├── sonrai-agent.md # API integration, brand alignment
├── documentation-agent.md # AWS-style documentation standards
├── kiroween-submission-agent.md # Deadline-focused prioritization
└── ... (9 more specialized agents)
Before vs After:
| Without Steering | With Steering |
|---|---|
| Generic code suggestions | Domain-specific patterns |
| Inconsistent architecture | Unified design language |
| Random documentation style | AWS-style standards |
| No performance awareness | 60 FPS requirement enforced |
4. 🔄 Hooks: Automated Quality Gates
7 Hooks Catching Problems Before Commit:
| Hook | Trigger | What It Does | Impact |
|---|---|---|---|
qa-review-src-changes |
File save | Runs relevant tests | 5s vs 30s feedback |
sync-arb-with-backlog |
Backlog edit | Updates quality scores | Automated tracking |
pre-commit-security |
Git commit | SAST + secret scan | Zero secrets leaked |
test-api-integration |
API changes | Validates Sonrai calls | Catches errors early |
generate-coverage-report |
Test run | Tracks coverage gaps | Identifies blind spots |
Measured Result: Test-debug cycle reduced from 30 seconds to 5 seconds.
5. 🔌 MCP: GitHub Without Leaving the IDE
Operations Used:
create_issue→ Backlog items become GitHub issueslist_commits→ Track development progressget_pull_request_status→ Monitor CI/CDsearch_code→ Find implementation patternscreate_branch→ Feature branch management
The Workflow:
Kiro: "Create issue for arcade mode timer"
↓
GitHub MCP creates issue #42
↓
Kiro implements feature
↓
GitHub MCP creates PR
↓
Kiro monitors CI/CD status
↓
GitHub MCP merges after tests pass
Impact: 40% reduction in context switching between IDE and browser.
🏛️ THE DIFFERENTIATOR: 12-Agent Architecture Review Board
What makes this submission unique:
Kiro conducted a comprehensive architecture review acting as 12 specialized agents:
| Agent | Domain | Key Contribution |
|---|---|---|
| 1. Product Manager | Planning | Sprint velocity, backlog prioritization |
| 2. Architecture Lead | Design | Patterns, code organization |
| 3. QA Lead | Testing | 667 tests, coverage analysis |
| 4. Security Lead | Security | SAST, vulnerability assessment |
| 5. Operations Lead | Reliability | Deployment, monitoring |
| 6. DevEx Lead | DX | Onboarding, tooling |
| 7. UX/Design Lead | UX | Accessibility, usability |
| 8. Documentation Lead | Docs | Standards, maintenance |
| 9. DevOps Lead | CI/CD | Branch management, automation |
| 10. Standards Lead | Quality | Code consistency |
| 11. Product Vision Lead | Strategy | Roadmap, market fit |
| 12. Kiroween Lead | Deadline | Submission optimization |
The Output:
- 47 recommendations across all domains
- 7.4/10 quality score with detailed breakdown
- 4-5 sprint improvement roadmap
- 160-200 hours of identified improvements
Why This Matters: This demonstrates Kiro as a complete development organization, not just autocomplete.
🚧 Challenges We Conquered
Challenge 1: Performance Bottleneck
| Problem | Solution | Result |
|---|---|---|
| 15 FPS with 100 zombies | Spatial grid collision ($O(n^2) \rightarrow O(n)$) | 60 FPS with 500+ entities |
The Math: $$T_{naive} = n \times m \times t_{check} \quad \text{(every projectile vs every zombie)}$$ $$T_{grid} = m \times t_{insert} + n \times k \times t_{check} \quad \text{(only nearby cells)}$$
Where $k$ = average entities per cell (typically 3-5), giving us 18.5× speedup.
Challenge 2: API Rate Limiting
| Problem | Solution | Result |
|---|---|---|
| Batch quarantine hitting limits | 10 calls/batch with 1s delays | Reliable batch operations |
Challenge 3: Deadline Pressure
| Problem | Solution | Result |
|---|---|---|
| 11 days to build production game | Kiro's spec-driven development | Zero P0 bugs, 667 tests |
🏆 Why We Should Win: Resurrection Category
Perfect Thematic Fit
| Resurrection Element | Our Implementation |
|---|---|
| Dead Technology | 8-bit/16-bit retro gaming aesthetic |
| Brought Back to Life | Modern cloud security education |
| Zombie Theme | Literal zombies = "dead" unused identities |
| Halloween Vibes | Spooky arcade action with real consequences |
Technical Excellence
✅ Real API Integration — Not mock data, actual Sonrai GraphQL calls ✅ Production Quality — 667 tests, 60 FPS, comprehensive error handling ✅ Professional Architecture — State machines, spatial grids, event systems
![]() WannaCry Boss |
![]() Heartbleed Boss |
![]() Scattered Spider Boss |
Boss battles teach real security breach history—WannaCry, Heartbleed, Scattered Spider
Business Value
✅ Market — $X billion security training gap ✅ Audience — 9-year-olds to CISOs ✅ Scalability — Platform adapts to any security concept ✅ Partnership — Built on Sonrai Security's Cloud Permissions Firewall
🎯 Judging Criteria Alignment
Potential Value (33%)
| Criteria | Our Evidence |
|---|---|
| Solves Real Problem | Cloud security training gap—abstract concepts made tangible |
| Widely Useful | Education, corporate training, conference demos, recruitment |
| Easy to Use | Game interface vs technical documentation |
| Accessible | 9-year-olds to CISOs can understand and enjoy |
Implementation (33%)
| Criteria | Our Evidence |
|---|---|
| Extensive Kiro Usage | All 5 features: vibe, specs, steering, hooks, MCP |
| Multiple Capabilities | 12-agent ARB demonstrates sophistication |
| Clear Documentation | 43 files in .kiro directory with evidence |
| Sophisticated Integration | Real API calls, not mock data |
Quality & Design (33%)
| Criteria | Our Evidence |
|---|---|
| Polished Aesthetic | Consistent retro 8-bit visual style |
| Performance | 60 FPS with 500+ entities (4× improvement) |
| Testing | 667 automated tests, 100% pass rate |
| Documentation | AWS-style standards, evidence-based claims |
📁 Evidence Location
Judges: Everything is in the .kiro/ directory!
.kiro/
├── steering/ # 15+ agent role definitions
│ ├── product-manager.md # 400+ lines
│ ├── architecture-agent.md # 350+ lines
│ ├── security-agent.md # 200+ lines
│ └── ... (12 more)
├── specs/ # 6 complete feature specs
│ ├── arcade-mode/
│ ├── jit-access-quest/
│ ├── service-protection-quest/
│ └── ... (3 more)
├── hooks/ # 7 automated workflows
├── evidence/ # Screenshots & recordings
│ ├── gameplay/ # 12 curated gameplay shots
│ ├── screenshots/ # 25+ raw captures
│ └── booth_photos/ # Photo booth composites
└── ARCHITECTURE_REVIEW_BOARD_REPORT.md # 12-agent review
🖼️ Visual Gallery
![]() AWS Org Lobby |
![]() Zombie Blasting |
![]() Purple Shields |
![]() Arcade Results |
![]() Production Outage |
![]() Photo Booth |
🚨 NEW: Production Outage Simulation
Production Outage mode—experience the chaos when zombies overwhelm your cloud infrastructure!
🎬 Try It Yourself
# Clone and setup (2 minutes)
git clone https://github.com/colehorsman/zombie_game.git
cd zombie_game
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env # Add Sonrai credentials if available
# Run the game
python3 src/main.py
# Run tests (667 passing)
pytest tests/ -v
🎮 Cheat Codes:
UNLOCK— Access all levelsGOD— Invincibility modeARCADE— Jump to arcade mode
📸 Photo Booth Feature
Take retro-filtered selfies with your game stats—shareable proof of your zombie-blasting skills!
🔗 Links
| Resource | URL |
|---|---|
| GitHub Repository | https://github.com/colehorsman/zombie_game |
| Demo Video | [YouTube - to be added] |
| Developer | Cole Horsman (cole.horsman@sonraisecurity.com) |
| Company | Sonrai Security |
🏁 Conclusion
Sonrai Zombie Blaster proves Kiro is more than autocomplete—it's a complete development organization.
| Role | What Kiro Did |
|---|---|
| Product Manager | 2 sprints, backlog, velocity tracking |
| Technical Lead | Architecture, 4× performance improvement |
| QA Engineer | 667 tests, 100% pass rate |
| Security Lead | SAST, zero secrets committed |
| Documentation Lead | 43 files, AWS standards |
| + 7 More Agents | 12-agent ARB with 47 recommendations |
Built in 11 days. 667 tests. 60 FPS. Zero P0 bugs.
This isn't autocomplete. This is AI-powered software development lifecycle management.
Built with Kiro. Powered by Sonrai. Ready for production. 🎮🧟🚀
#BuildWithKiro #Kiroween2025 #hookedonkiro
Built With
- amazon-web-services
- api
- identity
- kiro
- pygame
- python
- sast
- sonrai






Log in or sign up for Devpost to join the conversation.