GitHubGitHub RepoNeeds Review

huangruiteng/loopx

LoopX gives long-running coding-agent work durable goals, gates, ownership, evidence and quotas across bounded turns. It is useful when chat history and a scheduler no longer provide enough control.

GitHub

Open Source Open Markdown Open JSON

Source Summary

LoopX **v0.4.x** is a local, agent-agnostic state kernel for Codex, Claude Code, Cursor and custom runners. It preserves **goals, gates, todos, evidence and quota** while agents execute bounded turns; two published trajectories span **200+ elapsed hours** each.

Practical Implication

Use it when work crosses sessions or agents and ownership, stop conditions or human approvals must remain explicit. Its quota check, todo claims, validated writeback and review packets offer a concrete harness pattern even if you do not adopt the project.

Agent-Ready Context

LoopX **v0.4.x** is a local, agent-agnostic state kernel for Codex, Claude Code, Cursor and custom runners. It preserves **goals, gates, todos, evidence and quota** while agents execute bounded turns; two published trajectories span **200+ elapsed hours** each.

Use it when work crosses sessions or agents and ownership, stop conditions or human approvals must remain explicit. Its quota check, todo claims, validated writeback and review packets offer a concrete harness pattern even if you do not adopt the project.

It is early software, and the long trajectories measure wall-clock project lifetime rather than continuous execution or independent reproduction. **Production writes and final authority stay with the human**, so LoopX is not an autonomous production controller.

Connected Context · Feed7 Judgment

LoopX makes durable state, bounded turns, evidence, quotas, claims, and approval gates one local kernel for work spanning agents or sessions. Against the candidates, it strengthens the case that continuity and “done” belong in the harness rather than any model context, while preserving human authority. Its long trajectories demonstrate persistence, not autonomous execution, production safety, or independently reproduced reliability.

What Does Done Even Mean? Agents and Paperclip's Liveness Model - Dotta, PaperclipPaperclip defines completion through evidence, authority, residual risk, and next ownership; LoopX operationalizes closely related controls through gates, evidence, todo claims, review packets, and human final authority.Chained Recursive Language Models for Multi-Iteration ReasoningChained RLM preserves summaries, a blackboard, and artifacts across fresh contexts; LoopX supplies a model-agnostic external state kernel with additional ownership, quota, validation, and approval controls for that continuity problem.Why We Killed Our Multi-Agent Pipeline — Subbiah Sethuraman and Abhilash Asokan, ZS AssociatesThe failed fixed-agent chain shows why handoffs and fragmented ownership can damage coherence; LoopX addresses the coordination boundary with explicit claims and writeback, but does not overturn the candidate’s case for one agent retaining end-to-end ownership.The Factory That Dreams: 39 AI Agents, No Framework - Rushabh Doshi, MachinecraftThe factory system emphasizes organized memory, specialist roles, and human approval; LoopX provides a local implementation pattern for preserving shared state and approvals without claiming to replace human production authority.

Context Map

agentcodingresearch#harness-engineering#multi-agent#agent-memory

Uncertainty

It is early software, and the long trajectories measure wall-clock project lifetime rather than continuous execution or independent reproduction. **Production writes and final authority stay with the human**, so LoopX is not an autonomous production controller.