Show HN: Aura, an LLM coding harness that dogfooded itself
Aura is an LLM coding harness designed to enhance coding efficiency through a structured engineering process. It utilizes a two-agent system, the Planner and Worker, to create technical specifications and execute code edits while ensuring validation and recovery. The tool has been extensively tested on its own codebase, demonstrating its capabilities through significant API usage.
- ▪Aura transforms any model into a more effective coding tool by implementing a structured engineering loop.
- ▪The system consists of a Planner that creates specifications and a Worker that executes them with validation.
- ▪Aura has processed over 1.1 billion tokens and completed nearly 30,000 API requests while developing itself.
Opening excerpt (first ~120 words) tap to expand
Aura An LLM coding harness — turns any model into a better engineer through process, tools, context, validation, and recovery. Why Aura? Aura is an LLM coding harness. It takes your codebase, your prompt, and a capable model — then runs it through a real engineering loop: repo awareness → Planner spec → Worker execution → surgical edits → validation → recovery → final receipt. The Planner reads your code, understands the project structure, and writes a precise technical specification. You review (and edit) the spec before it reaches the Worker. The Worker executes the spec with read/write filesystem access, runs validation commands, and reports back a summary. Every file write shows a diff before it touches disk. Backups and git commits make experiments reversible.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at GitHub.