alisonjieli-png

Loop Engine — Development skill for Claude Code

Development community

Baltor: a reviewed library of skills, AGENTS.md sections, subagents, commands, hooks and protocol server settings that coding agents search and fetch for each step.

How to install Loop Engine

This entry records only its repository, not the path inside it, so there is no exact command to give. Open alisonjieli-png/loop-engine and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Loop Engine does

Baltor: a reviewed library of skills, AGENTS.md sections, subagents, commands, hooks and protocol server settings that coding agents search and fetch for each step. This repository holds Loop Engine, its open-source engine.

Alternatives in Development

Pubmed MCP Server — Search PubMed/Europe PMC, fetch articles and full text (PMC/EPMC/Unpaywall), citations, MeSH terms via MCP 150 ★
Pi — PI Wisdom-in-Action Engine v23 — top mandates, scene router, delivery quality gates, anti-patterns, battle mom 107 ★
Search Prior Art — Conduct comprehensive prior art search using 7-step methodology across 100M+ patents via BigQuery 91 ★

README

Baltor

Harness and agent optimized operation.

Baltor makes coding harnesses and multi-agent systems as efficient as possible. The aim is a frontier harness, multi-agent system or fabric that can take on any unseen task in the most efficient way:

the exact context each step needs, and no more;
code that already exists is reused instead of written again;
the right amount of intelligence and heuristics for each decision;
small models doing much more, including work that runs overnight.

The task can be anything: a data pipeline, research, a data cleanup or a piece of software. Turning a complex problem into a solution you can run again is one benefit of this work. It is not the goal itself.

Baltor is the product and the website at . Loop Engine is this repository: the open engine behind Baltor, the Python package `loop-engine` and the `loop-engine` command.

One harness for each step

A harness is the program that runs a coding agent, such as OpenCode, Codex, Claude Code or Pi. Baltor is designed to start a fresh harness for every step of a task. Each one holds only the context, skills, tools and code that its small step needs, so no step carries the context rot of a long, overloaded session.

A task
├── Step 1: a fresh harness with the material step 1 selected
├── Step 2: a fresh harness with the material step 2 selected
├── Step 3: a fresh harness that reuses tested code instead of writing it
└── Check: an independent step accepts or rejects the result

Inside the engine, each step is a discrete cognitive or act step Loop node: one independently governed instance of the Loop runtime, responsible for one clearly defined cognitive step or action, with its own contracts, permissions and history. Read the [complete behavioral explanation](ASTRA.md#complete-behavioral-explanation) before describing one.