agent-teams-coder

Agent Teams Coder

A multi-agent software development collaboration framework powered by Claude Code. Seven specialized AI agents work together under a strict shared memory governance model.

Installation

Option A: Claude Code Plugin (Recommended)

# Step 1: Add the marketplace
claude plugin marketplace add Xinjing-Guo/Agent-Teams-Coder --sparse .claude-plugin plugin

# Step 2: Install the plugin
claude plugin install agent-teams-coder

Then in any Claude Code session:

/agent-team Write a high-performance FFT library in Python and C

/agent-team auto-detects tmux:

• tmux available → asks you: multi-window (7 visible panes) or single-window (automated)
• tmux not available → runs in single-window mode automatically

Option B: tmux Multi-Pane (No Plugin Needed)

git clone https://github.com/Xinjing-Guo/Agent-Teams-Coder.git
cd Agent-Teams-Coder
./panel.sh

Choose a launch preset:

Option	Agents	Use Case
a) Full team	All 7	Complete workflow
b) Core dev	Marshall + Euler + Forge + Sentinel	Algorithm → Code → Test
c) Leader only	Marshall	Planning and coordination
d) Algorithm + Dev	Euler + Forge	Focused implementation
e) Test + Analysis + Docs	Sentinel + Lens + Atlas	Post-development pipeline

Each agent supports model selection:

./start-euler.sh           # Default: Sonnet
./start-euler.sh opus      # Opus for complex algorithm design
./start-chronicle.sh haiku # Haiku for logging (cost-efficient)

Option C: Local Plugin Install

git clone https://github.com/Xinjing-Guo/Agent-Teams-Coder.git
claude plugin add ./Agent-Teams-Coder/plugin/agent-teams-coder

Plugin Management

claude plugin list                          # List installed plugins
claude plugin enable agent-teams-coder      # Enable
claude plugin disable agent-teams-coder     # Disable
claude plugin uninstall agent-teams-coder   # Uninstall

---

Team Members

Codename	Role	Skills (32 total)
Marshall	Leader	task-decomposition, risk-assessment, progress-tracking, team-coordination
Euler	Algorithm Designer	algorithm-design, complexity-analysis, numerical-methods, optimization-algorithms, data-structures, statistical-modeling
Forge	Code Developer	multi-language-coding, code-review-checklist, python-expert, c-cpp-expert, r-julia-expert, build-and-packaging
Sentinel	Code Tester	test-strategy, bug-tracking, python-testing, c-cpp-testing, performance-testing
Lens	Code Analyst	code-analysis-framework, static-analysis, design-pattern-recognition, call-graph-generation
Atlas	Documentation Engineer	manual-structure, api-documentation, tutorial-writing, diagram-generation
Chronicle	Log Recorder	activity-logging, changelog-generation, decision-log

Standard Workflow

Phase 1: Requirements   → Marshall decomposes requirements, Chronicle starts logging
Phase 2: Algorithm       → Euler designs algorithm, aligns with Forge          → CC Chronicle
Phase 3: Development     → Forge implements code based on Euler's algorithm    → CC Chronicle
Phase 4: Testing         → Sentinel tests rigorously, broadcasts test report   → CC Chronicle
                            ↻ Bug found → Forge fixes → CC Chronicle → Sentinel retests
Phase 5: Analysis        → Lens analyzes code structure, line-by-line          → CC Chronicle
Phase 6: Documentation   → Atlas integrates manual (test cases + code analysis)→ CC Chronicle
Phase 7: Delivery        → Marshall consolidates, Chronicle generates summary  → CC Chronicle

Core Mechanisms

1. Shared Memory with Approval Governance

The shared memory (shared-memory.json) stores team-wide architectural decisions, API conventions, and coding standards. It is protected by an approval mechanism: