attractor
nicheNLSpec blueprint for building a non-interactive coding agent; clever but unproven.
What it is
A set of markdown specifications (NLSpecs) for constructing a non-interactive coding agent suitable for autonomous use in a software factory. The user prompts a coding agent (e.g., Claude Code) to implement the Attractor based on these specs.
How it differs from vanilla Claude
Vanilla coding agents require human interaction for each task. Attractor aims for a fully automated loop that reads task specifications and produces code without human guidance. The specs define the structure, but the baseline agent (e.g., Claude Code) must still write the code—Attractor is a pattern, not a tool.
Skill, plugin, or workflow shift?
Three spec files (~50KB total). Implementation scope includes agent loop orchestration, LLM client abstraction, and task execution pipeline.
Devil's advocate — is this just complexity?
This is a spec that says 'build a non-interactive coding agent using an existing coding agent'. A capable base agent (Claude Code with proper prompting) can already run in non-interactive mode by passing task definitions as input—no custom agent loop needed. The specs add complexity by inventing a new abstraction layer (NLSpecs) that duplicates what structured prompts already achieve. The project is essentially a meta-prompt: 'implement this spec'. The value is not in the spec itself but in whatever implementation you build from it, which you could have built directly. It's a solution looking for a problem.
What would make it better
Provide a reference implementation (e.g., Python package) that users can install and run immediately, with clear benchmarks showing performance improvement over a simple non-interactive prompt. Include integration tests and a real-world software factory example (e.g., GitHub Action pipeline).
The honest case for it
By formally specifying the agent loop and LLM client interface, Attractor enables reproducible, testable, and evolvable agent architectures. It decouples the 'what' (spec) from the 'how' (implementation), allowing teams to swap LLM backends or loop logic without rewriting prompts. This discipline is essential for production software factory automation where reliability and traceability matter.
Who it's for
Audience fit
Depth and leverage for a technical engineer who wants to understand it and level up their workflow — not just offload work.
Value for someone who wants a more capable tool without the technical depth — accessible, does-it-for-you.
Requires deep understanding of agent loops and willingness to implement from specs; offers automation leverage but demands significant setup.