Skip to main content
Score monitors your GitHub repository, picks up labeled issues, dispatches AI agents to implement the work, gates every PR with make verify, and merges when green — all without you lifting a finger after the initial setup.

Introduction

Understand what Score is, how its four-phase daemon works, and when it’s the right tool for your workflow.

Quickstart

Get Score running and watch it dispatch its first agent and land a PR in under 10 minutes.

CLI Reference

Every command, flag, and subcommand — score up, score repair, score tui, and more.

Guides

Deep-dives on managed mode, agent runtimes, the make verify contract, and fleet operations.

​
How It Works

Score runs a continuous poll loop. Each tick it works through four phases in order — cleanup, dispatch, landing, and repair — ensuring work is always progressing and nothing gets stuck.
1

Label a GitHub issue

Add an epic:-prefixed label to any GitHub issue in your repository. That’s the only signal Score needs to know an issue is ready for an agent.
2

Score dispatches an agent

On the next tick, Score creates a git worktree for the issue, writes a task briefing, and spawns an AI agent session (Claude via tmux, or opencode) to implement the work.
3

The agent pushes a PR

The agent works inside the isolated worktree, commits its changes, and opens a pull request. Score tracks the PR each tick — soaking it for consecutive green make verify runs before it qualifies for merge.
4

Score merges when green

Once the PR passes verification for the configured number of soak ticks and carries no blocking labels, Score merges it and reclaims the worktree — ready for the next issue.

​
Key Features

Autonomous Pipeline

Score closes the full loop — from labeled issue to merged PR — with no human steps in between. You set the policy; Score executes it.

Agent Runtimes

Choose between the claude harness (tmux sessions) and opencode (a durable HTTP session per issue). Both are first-class; switch per project in config.

Managed Mode

score up registers Score as a supervised service under launchd (macOS) or systemd (Linux), so it survives reboots and restarts automatically on failure.

TUI Monitor

score tui opens a live terminal UI showing all projects, recent merges, daemon telemetry, and log history — no browser required.

Dry-Run Mode

Pass --dry-run to preview every phase decision — what would be dispatched, soaked, merged, or repaired — without touching a session, worktree, or branch.

make verify Contract

Score carries zero language assumptions. Every managed repo exposes make verify; Score runs it before any merge. Bun, Go, Rust — all identical to Score.