Design a standardized, shareable format for describing multi-phase AI agent workflows.
Publish and version-pin reusable agent loop definitions through an OCI-compliant registry.
Build a client tool that reads, validates, and executes Agent Loop YAML documents.
Track exact dependency versions of agent loops used in a project via a lock file.
| stumpyfr/loop-spec | 0xkinno/neuralvault | 0xlocker/d17-contracts | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | — | TypeScript | Solidity |
| Setup difficulty | easy | hard | hard |
| Complexity | 2/5 | 4/5 | 5/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
There is nothing to install, only a specification document to read and implement against.
loop-spec is a written specification document, not a runnable tool, that defines a standard way to describe and share reusable instructions for AI agents, called Agent Loops. It is currently a draft, version 0.1.0. An Agent Loop is a YAML file that lays out how an AI agent should carry out a piece of work: what information it needs as input, what stages or phases the work moves through, what actions happen in each phase, how the agent knows a phase is finished, what it produces as output, how it moves from one phase to the next, and when it should stop and ask a human for a decision instead of continuing on its own. The bulk of the document is about how to package and share these loop files using OCI Artifacts, the same general packaging system used for container images, but applied here to plain YAML files instead of software containers. This means loop definitions can be published to any registry that already supports that packaging format, given a fixed version identifier so the exact same loop can be pulled again later, optionally signed to prove who published them, and organized into named collections. The spec also defines a project level file for listing which loops a project depends on, plus a lock file that records the specific version of each one actually in use, similar to how some programming tools track exact dependency versions. The author explains this is closely modeled on an existing, separate specification for something called Agent Skills, following the same overall pattern but applied to a different kind of building block: a skill teaches an agent how to do a type of task, while a loop is described as directing a full bounded run through defined stages with a beginning and an end. The two are meant to work side by side rather than replace one another. The document notes that a command line tool called agentkit implements this specification, but states plainly that the specification itself, not any particular tool, is the authoritative source of truth. It is released under the Apache 2.0 license, which permits free use, including commercial use, provided the license terms are followed.
A draft specification defining a portable YAML format and OCI-based packaging system for sharing reusable, phase-based AI agent orchestration instructions.
Use freely, including commercial use, as long as you follow the Apache 2.0 license's terms, such as keeping attribution notices.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.