# Aura Agent

An autonomous task orchestrator for long-running projects that decomposes a written goal into subtasks, routes them to isolated worker agents, and verifies results against acceptance criteria.

Aura reads a task file describing a goal and continuously decomposes it into concrete subtasks, dispatching each to one of two worker roles, a stronger Pro worker and a faster Flash worker, which can run backed by Claude Code, Qwen Code, Codex CLI, ds_code or OpenCode. Work can run inside isolated Docker containers by default, or locally for a single session, and editing the task file while Aura is running causes it to detect the change, re-plan and continue automatically.

A Rich-based terminal dashboard shows progress by default, with a plain command-line mode available, and a worker-delegation mode can be set per run to control whether task assignment is automatic or biased toward one worker role. It installs as an editable Python package and includes an interactive setup command for configuring API keys and models.

Website: https://github.com/erickong/aura-agent