Skip to main content

Specification-driven AI development system with multi-agent orchestration

Project description

IXV-Agents

Overview

IXV-Agents is a specification-driven AI development system that organizes multiple AI agents into a fixed, role-based team. It integrates agile roles and events with specification-driven development to ensure governance, traceability, and practical enterprise usage.


Core Concept

Fixed roles, evolving skills. Humans define intent and specifications, while AI agents collaborate as a structured team.

  • Specifications (Specs) are the single source of truth
  • Roles define responsibility boundaries
  • Events establish development rhythm

4 Principles

  1. Specs are living documents
  2. Specs are the Single Source of Truth (SSoT)
  3. Change and iteration are assumed
  4. AI reduces cost, humans decide

7 Processes

# Process Output Approval
1 Constitution CONSTITUTION.md Human
2 Specify README.md (SSoT) Human
3 Plan docs/* Human(*)
4 Tasks queue/tasks/, dashboard.md -
5 Implement Code + Tests, reports/*.yaml -
6 Verify/Accept dashboard.md, Backlog update Human
7 Migration/Op → Process 2 or 4 -

(*) = when needed

See templates/PROCESS.md for details.


Agent Team Composition (Fixed)

Role Count Responsibility
Product Owner (PO) 1 Define goals and priorities, create specifications
Scrum Master (SM) 1 Orchestrate workflow, decompose and assign tasks
Development (Dev) 3 Implementation

Prerequisites


Setup

AI Editor

Install one of the following.

OpenCode (Default)

Claude Code

Terminal Multiplexer

macOS: tmux

  • Command install: brew install tmux
  • Other installation methods: See official Wiki

Windows: psmux (tmux-compatible)

  • Command install: irm https://raw.githubusercontent.com/marlocarlo/psmux/master/scripts/install.ps1 | iex
  • Other installation methods: See official repository
  • PowerShell 7+ required

Usage

1. Start Agents

macOS:

# Start with OpenCode (default)
./scripts/boot.sh

# Start with Claude Code
./scripts/boot.sh --claude-code

# Specify model
./scripts/boot.sh --model anthropic/claude-opus-4-5

Windows (PowerShell):

# Start with OpenCode (default)
.\scripts\boot.ps1

# Start with Claude Code
.\scripts\boot.ps1 -ClaudeCode

# Specify model
.\scripts\boot.ps1 -Model anthropic/claude-opus-4-5

On first run, the workspace is automatically initialized.

2. Session Layout

A single tmux session (ixv-agents) is created and automatically attached:

[ixv-agents] All Agents (5 panes)
┌─────────┬───────┬───────┬───────┐
│   PO    │ Dev1  │ Dev2  │ Dev3  │
│  (0.0)  │ (0.2) │ (0.3) │ (0.4) │
├─────────┤       │       │       │
│   SM    │       │       │       │
│  (0.1)  │       │       │       │
└─────────┴───────┴───────┴───────┘

Usage:

  • Communicate your requirements to PO (upper-left pane), and tasks will be assigned to the Dev team via SM
  • Other panes (SM, Dev1-3) operate automatically; no manual interaction required

Detach from session:

  • Ctrl+b d to detach (session continues in background)

Reattach to session:

tmux attach-session -t ixv-agents

3. Stop Sessions

# macOS
./scripts/stop.sh
./scripts/stop.sh --force    # Force stop if processes remain
# Windows
.\scripts\stop.ps1
.\scripts\stop.ps1 -Force    # Force stop if processes remain

4. Setup New Workspace

macOS:

./scripts/setup_workspace.sh

# Skip backup and reinitialize only
./scripts/setup_workspace.sh --no-backup

Windows (PowerShell):

.\scripts\setup_workspace.ps1

# Skip backup and reinitialize only
.\scripts\setup_workspace.ps1 -NoBackup

If an existing workspace/ exists, it will be backed up to backups/, and a new workspace will be created.

tmux Quick Reference

Action Command
Detach from session Ctrl+b d
Reattach to session tmux attach-session -t ixv-agents
List sessions tmux ls

Directory Structure

ixv-agents/
├── roles/              # Role instructions (PO, SM, Dev)
├── skills/             # AI CLI skill definitions
├── templates/          # Workspace initialization templates
│   └── queue/          # Queue and report templates
├── scripts/            # Startup and management scripts
│   ├── banner.sh / .ps1           # Display banner
│   ├── boot.sh / .ps1             # Start agents
│   ├── stop.sh / .ps1             # Stop agents
│   ├── setup_workspace.sh / .ps1  # Initialize workspace
│   └── tmux-help.txt              # In-pane help text
├── OLD/                # Legacy assets (kept for reference)
├── backups/            # Workspace backups [.gitignore]
├── workspace/          # AI editor working directory [.gitignore]
├── docs/               # Documentation
├── Spec.md             # System specification
└── README.md

workspace/ Directory

workspace/ is the directory where AI editors actually perform their work. It is isolated from the repository root, preventing AI editors from accessing tool READMEs and other unrelated files.

workspace/
├── README.md           # Project spec (Single Source of Truth)
├── CONSTITUTION.md     # Project constitution
├── PROCESS.md          # Process and operations
├── AGENTS.md           # AI conduct guidelines
├── roles -> ../roles   (symlink)
├── .claude/skills -> ../../skills   (symlink)
├── .opencode/skills -> ../../skills (symlink)
├── queue/              # Inter-agent communication
│   ├── dashboard.md    # Project status board
│   ├── po_to_sm.yaml   # PO -> SM
│   ├── tasks/          # SM -> Dev
│   └── reports/        # Dev -> SM
└── (artifacts)         # Implementation code, tests, etc.

Operational Principles

  • Single Source of Truth: Always reference workspace/README.md
  • Traceability: Track via spec_ref / request_id / task_id
  • Role Boundaries: Writing to files outside role scope is prohibited

Key Documents

  • Spec.md: System architecture, roles, workflow, and constraints
  • docs/20260129implementation-plan.md: Implementation plan
  • docs/20260201directory-restructure-plan.md: Directory restructure plan

License

TBD

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ixv_agents-0.0.1.tar.gz (612.4 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

ixv_agents-0.0.1-py3-none-any.whl (4.1 kB view details)

Uploaded Python 3

File details

Details for the file ixv_agents-0.0.1.tar.gz.

File metadata

  • Download URL: ixv_agents-0.0.1.tar.gz
  • Upload date:
  • Size: 612.4 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.9.6

File hashes

Hashes for ixv_agents-0.0.1.tar.gz
Algorithm Hash digest
SHA256 e780f62e14a02bfb36ba1fb8a1b7218ed125524151405527d5b39fbf45887e0f
MD5 a09e65bdd5905d830b6d9be9d5c0ae72
BLAKE2b-256 adee1ecf966f81c7b807995e0dacddf9a03587e5959be683dd675aa605ed39e8

See more details on using hashes here.

File details

Details for the file ixv_agents-0.0.1-py3-none-any.whl.

File metadata

  • Download URL: ixv_agents-0.0.1-py3-none-any.whl
  • Upload date:
  • Size: 4.1 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.9.6

File hashes

Hashes for ixv_agents-0.0.1-py3-none-any.whl
Algorithm Hash digest
SHA256 a02b74bc33f29cc81024763580439c7b14046c266c87a63b9844f744799c130b
MD5 87add1ac768470b0ff9d70cde4b0e4b9
BLAKE2b-256 c15fa0811b81493e44e722097ce8966f860f5d83b6afd397f88108440205e7c5

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page