A multi-agent development system themed after the Three Kingdoms era, with 12 specialized AI agents for software development workflows.
Supports both Kiro CLI and Claude Code.
| Feature | Kiro CLI | Claude Code |
|---|---|---|
| Agents | 12 JSON files | 12 Markdown files |
| Steering / Memory | steering/*.md (always-on) |
CLAUDE.md (always-on) |
| Skills | 10 skill directories | 10 skill directories |
| Subagent tool | use_subagent |
Task |
| Install path | ~/.kiro/ |
~/.claude/ |
git clone https://github.com/frankekn/three-kingdoms-kiro.git
cd three-kingdoms-kiro
./install.shInteractive menu lets you choose:
- [1] Kiro CLI only
- [2] Claude Code only
- [3] Both
Or use CLI flags:
./install.sh --kiro # Kiro CLI only
./install.sh --claude-code # Claude Code only
./install.sh --both # Both platforms./uninstall.sh # Interactive menu
./uninstall.sh --both # Remove from both platforms| Chinese | Agent Name | Role |
|---|---|---|
| Zhuge Liang / 諸葛亮 | zhuge | Strategy, orchestration, task delegation |
| Zhao Yun / 趙雲 | zhaoyun | Core code implementation |
| Guan Yu / 關羽 | guanyu | Staff-level code review |
| Zhang Fei / 張飛 | zhangfei | Bug hunting, stress testing |
| Zhou Yu / 周瑜 | zhouyu | UI/UX engineering |
| Xiao Qiao / 小喬 | xiaoqiao | Visual design, copy, micro-interactions |
| Cao Cao / 曹操 | caocao | CI/CD, automation, deployment |
| Pang Tong / 龐統 | pangtong | Architecture audit |
| Guo Jia / 郭嘉 | guojia | Research, algorithms, models |
| Xun Yu / 荀彧 | xunyu | Task integration, breakdown |
| Huang Zhong / 黃忠 | huangzhong | Staff review gatekeeper |
| Lu Su / 魯肅 | lusu | UX flow, information architecture |
Zhuge Liang plans -> Guo Jia researches -> Pang Tong audits -> Xun Yu breaks down tasks -> Zhao Yun + Zhou Yu implement (TDD) -> Guan Yu + Huang Zhong review -> Cao Cao deploys
Zhuge Liang locates -> Zhang Fei hunts -> Zhao Yun fixes (TDD) -> Guan Yu reviews
Pang Tong audits architecture -> Zhuge Liang plans -> Zhao Yun refactors -> Guan Yu + Huang Zhong review
Huang Zhong + Guan Yu review -> Pang Tong audits architecture -> Zhao Yun fixes
| Skill | Description |
|---|---|
| verify-pipeline | Build, type check, lint, test, security scan |
| tdd-flow | Test-driven development workflow |
| security-audit | Security review checklist |
| search-before-code | Search existing solutions before writing new code |
| refactor-clean | Dead code removal, dependency cleanup |
| deploy-checklist | Deployment verification checklist |
| prd-template | Product requirements document template |
| design-template | System design document template |
| task-template | Task breakdown template |
| review-template | Code review report template |
When agents are chained in a workflow, each agent ends its response with a structured handoff block. This keeps inter-agent context compact and prevents context window overflow.
---HANDOFF---
outcome: [one sentence summary]
files_changed: [file paths, max 10]
decisions: [max 3 bullets]
risks: [max 2 bullets]
next: [what the next agent should focus on]
---END---
Rules:
- Handoff must be under 200 words
- Code belongs in files, not in the handoff
- Zhuge Liang (orchestrator) extracts only the handoff block when relaying context to the next agent
- When more than 5 handoffs accumulate, only the first and last 3 are kept
Automated checks on agent lifecycle events. No context window cost.
| Agent | Event | Action |
|---|---|---|
| Zhao Yun | stop | Verify pipeline reminder |
| Guan Yu | agentSpawn | Read diff first |
| Zhang Fei | preToolUse (shell) | Caution before destructive commands |
| Cao Cao | preToolUse (shell) | Dry-run reminder |
| Zhuge Liang | stop | Progress check |
three-kingdoms-kiro/
kiro/ # Kiro CLI files
agents/*.json # 12 agent definitions
steering/*.md # Always-on context
skills/*/SKILL.md # 10 reusable skills
claude-code/ # Claude Code files
agents/*.md # 12 agent definitions
CLAUDE.md # Always-on context
skills/*/SKILL.md # 10 reusable skills
install.sh # Interactive installer
uninstall.sh # Interactive uninstaller
MIT