A single Go binary that turns a solo developer into a full scrum team. ODA orchestrates AI coding agents via opencode to analyze, plan, implement, test, and review code — all driven by a lean scrum process.
ODA acts as your Scrum Master + Product Owner:
- You describe an idea or epic
- ODA breaks it into tasks (GitHub Issues)
- ODA plans a sprint and assigns tasks
- Workers (goroutines) process tasks through a pipeline: analysis → planning → plan review → coding → testing → code review → merge
- Each pipeline stage runs in a separate opencode session with a configurable LLM
- Results appear on a real-time HTMX dashboard
ODA manages GitHub Projects (board, columns) on your behalf. The default gh token does not include project scopes. Add them once:
gh auth refresh -s projectRequired scopes:
| Scope | Purpose |
|---|---|
repo |
Read/write issues, PRs, labels, milestones |
project |
Read/write GitHub Projects v2 (board, columns) |
# Install
go install github.com/crazy-goat/one-dev-army@latest
# Initialize in your project
cd /path/to/your/project
oda init
# Start the agent
odaODA stores its config in .oda/config.yaml:
github:
repo: "owner/repo"
use_projects: false # Set to true to enable GitHub Projects integration
dashboard:
port: 5000
workers:
count: 3
opencode:
url: "http://localhost:5002"
tools:
lint_cmd: "make lint"
test_cmd: "make test"
e2e_cmd: "make e2e"
pipeline:
stages:
- name: analysis
llm: claude-sonnet-4
- name: planning
llm: claude-opus-4
- name: plan-review
llm: claude-opus-4
- name: coding
llm: claude-sonnet-4
- name: testing
llm: claude-sonnet-4
- name: code-review
llm: claude-opus-4
- name: merge
manual_approval: true
max_retries: 5
sprint:
tasks_per_sprint: 10By default, ODA uses a label-based workflow and does not require GitHub Projects. To enable GitHub Projects integration:
- Set
use_projects: truein your config - Run
gh auth refresh -s projectto add project scope - Restart ODA
The label-based system (default) is recommended for new projects.
- Workers — goroutines with dedicated git worktrees for parallel task execution
- Pipeline — configurable stage machine with retry logic (max 5 retries, then escalate to user)
- GitHub — source of truth (issues, PRs, project board, milestones)
- SQLite — local metrics storage (tokens, costs, duration)
- Dashboard — HTMX + Go templates, real-time worker status
- Workflow: docs/workflow.md - Visual guide to the ODA ticket lifecycle
- Design: docs/plans/ - Architecture and implementation plans
- Status: Early development
MIT