Symphony Go

Symphony Go

A Go implementation of the [Symphony specification](../SPEC.md) — a long-running automation service that reads work from issue trackers, creates isolated workspaces, and runs AI coding agents for each issue.

Symphony supports two agent backends (**Gemini CLI** and **Claude Code**) and two issue trackers (**Linear** and **Jira Cloud**). Mix and match per workflow.

Architecture


Agent Backends

Symphony supports two agent backends. Set the `backend` field in your WORKFLOW.md to choose:

	Gemini CLI	Claude Code
Config key	backend: gemini (default)	backend: claude
Protocol	ACP — JSON-RPC 2.0 over stdio (long-running process)	NDJSON stream — one CLI invocation per turn
Session model	Single process, session persists in-memory	--resume across invocations, persisted to .symphony-session-id
Tool access	Client-side injection (ACP fs/terminal requests)	MCP servers (configured externally via .mcp.json or user config)
Permission handling	ACP session/request_permission auto-approve	--permission-mode bypassPermissions flag
Default model	gemini-3.1-pro-preview	claude-sonnet-4-6
TTY requirement	None	Requires pseudo-TTY (script -q /dev/null wrapper, handled automatically)

Gemini CLI Setup

npm install -g @google/gemini-cli
gemini auth login

For Linear integration, install the MCP extension:

gemini extensions install @google/mcp-linear

Claude Code Setup

npm install -g @anthropic-ai/claude-code

For Linear integration, add the MCP server globally:

claude mcp add -s user --transport http linear-server https://mcp.linear.app/mcp

This makes the Linear MCP server available in all workspaces. Alternatively, write a `.mcp.json` file in the workspace via the `after_create` hook for a self-contained setup.

Issue T