Symphony

Appearance

Quick Start ​

This guide walks you through creating your first issue and watching Symphony's agents work on it.

Project Auto-Detection ​

When Symphony starts, it automatically scans your workspace for projects. Each detected project gets:

  • A unique identifier prefix (e.g., SYM, API)
  • Its own kanban board
  • Independent agent configuration

Navigate to your project's board in the web UI at http://localhost:3333.

Creating an Issue ​

  1. Click the + New Issue button on the kanban board
  2. Fill in the issue details:
    • Title — A clear, concise description of the task
    • Description — Detailed requirements, acceptance criteria, and context
    • Priority — 1 (highest) to 5 (lowest)
    • Size — small, medium, large, or xlarge
  3. Click Create

The issue appears in the Backlog column.

Moving to Todo ​

Drag the issue from Backlog to Todo, or click the issue and change its status. Issues in todo are eligible for agent dispatch.

What Happens Next ​

Once an issue is in todo, the orchestrator picks it up based on priority and available agent slots:

  1. Phase Pipeline (if enabled) — The issue flows through research, architecture, and grooming phases before implementation
  2. Worker Dispatch — A worker agent is dispatched to an isolated git worktree
  3. Implementation — The agent writes code, runs tests, and commits changes
  4. PR Creation — The orchestrator auto-creates a pull request from the agent's changes
  5. Judge Review — A judge agent reviews the PR and approves or rejects it
  6. Human Merge — You review the approved PR and merge it

Monitoring Progress ​

Watch the agent's progress in real-time:

  • Kanban Board — Issues move across columns as they progress
  • Issue Detail — Click an issue to see agent findings, learnings, and PR status
  • Orchestrator Logs — Check the orchestrator terminal for dispatch and completion events

Tips ​

  • Start small — Create a well-scoped issue with clear requirements for your first run
  • Be specific — Agents perform best with detailed descriptions and concrete acceptance criteria
  • Watch the logs — The orchestrator terminal shows what's happening in real-time
  • Check findings — Agents record findings and learnings as they work, visible on the issue detail page