Analysis updated 2026-05-18
Define a portable folder format for an AI agent's instructions, skills, and Claude adapter files.
Compose several independent agent packages, like a product manager and two engineers, into one team with a shared channel.
Validate agent and team manifests against JSON Schema fixtures before shipping them.
| jackwener/openteamformat | bluejacketblackhawk/saysomething | chinaran0/wujing_dic | |
|---|---|---|---|
| Stars | 18 | 18 | 18 |
| Language | JavaScript | JavaScript | JavaScript |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 3/5 | 3/5 | 2/5 |
| Audience | developer | general | ops devops |
Figures from each repo's GitHub metadata at analysis time.
Requires npm install and understanding the YAML manifest schema before authoring a valid agent package.
OpenTeamFormat is a proposed file format for describing AI agent teams in a portable, shareable way. Instead of each tool inventing its own way to store an agent's instructions and skills, this project defines a standard folder layout that any Claude compatible tool can read. The smallest unit is a single agent package. It bundles a machine readable manifest file, a plain instruction file that tells the model how to behave, a Claude specific adapter file that points back to those instructions, and a folder of skills, which are small reusable capabilities the agent can use. The project also includes a team example showing how three separate agent packages, a product manager, a frontend engineer, and a backend engineer, can be composed into one team with a shared private channel. Two kinds of manifest exist. An OpenAgent manifest describes a single portable agent, including which local skills it owns and which external skills it references from other repositories. External skill references must point to a specific commit or tag rather than a moving branch, so that importing the same reference always produces the same result, using a branch name like main is allowed but flagged as a warning, or rejected outright under strict checking. An OpenTeam manifest describes a bundle of agents and the channels they belong to, referencing each agent by name rather than by a separate identifier. The format deliberately leaves out things like secrets, login sessions, running process state, message history, and file attachments, keeping the scope limited to the static description of who an agent is and what it can do. To help tool builders check that a package or manifest is well formed, the project ships JSON Schema files plus a set of example fixtures that are meant to pass validation, produce warnings, or fail outright. Running the included test suite validates both the shape of the files and rules specific to this format, such as making sure referenced skills and agents actually exist. The project is released under the Apache License 2.0.
A portable folder and manifest format for packaging AI agent instructions, skills, and Claude adapters so agent teams can be shared between tools.
Mainly JavaScript. The stack also includes JavaScript, YAML, JSON Schema.
Apache 2.0: use, modify, and distribute freely, including commercially, with patent protection and required attribution.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.