# Harness Library

> An open directory of reusable AI agent harnesses.

An agent harness is the surrounding setup that lets an AI model act on a task. It can include instructions, skills, tools, environment mappings, checks and workflows. Harness Library focuses on reusable, task-specific setups: from a small skill folder to an MCP workflow or a larger harness package.

New submissions are welcome. We’re looking for practical harnesses people can actually install and use.

- [Smooth Motion Graphics](https://openrootdev.github.io/harness-library/harnesses/smooth-motion-graphics/index.md): A compact skill for calm, clear brand animations.
- [ViperMesh for Blender](https://openrootdev.github.io/harness-library/harnesses/vipermesh-blender/index.md): A local Blender MCP workflow for building and checking 3D scenes.

Submit: https://openrootdev.github.io/harness-library/submit/index.md
About: https://openrootdev.github.io/harness-library/about/index.md
Machine-readable catalog: https://openrootdev.github.io/harness-library/catalog.json


# Submit an agent harness to Harness Library


> New submissions are welcome. We’re looking for practical harnesses people can actually install and use.


Canonical page: https://openrootdev.github.io/harness-library/submit/


Submission form: https://github.com/OpenRootdev/harness-library/issues/new?template=submit-harness.yml

No fork or pull request is required. A GitHub account is needed for the form.


## 1. Share your repository

Open the short GitHub form with a name, repository link and the task your harness helps an agent do. No fork or pull request is needed.


## 2. Explain how to install it

Add the focused package or harness branch, version and license, then list the install steps and anything people must connect or configure manually.


## 3. Get reviewed and listed

Maintainers check the setup and distribution scope, ask questions in your issue if needed, and publish approved listings. You can update the listing later.


## Before you submit

Bring repository, purpose/category, focused distribution/version/license, install steps and manual setup notes. Component counts and test evidence are optional. Harneloop is optional.


## Can I submit just a few skills?

Yes. A useful skill folder, MCP workflow or complete harness package can qualify. Explain the task, installation and manual setup clearly.


## Do I need to use Harneloop?

No. Submissions can use any agent or framework. Built with Harneloop is an optional attribution, reviewed separately from the listing.


## What if my repository contains a whole application?

Provide a focused package or a dedicated harness branch so users can install only the harness. Do not make them download unrelated videos, caches or application assets.


## Does a listing mean it works with every model?

No. Each page states its review scope, evidence and manual prerequisites. Setup reviewed and artifacts checked are different claims; neither guarantees every platform or model.


## Is submitting free?

Yes. Browsing and submitting to the directory are free. You need a GitHub account to open a submission issue; a Harness Library account is not required.


# About Harness Library

An agent harness is the surrounding setup that lets an AI model act on a task. It can include instructions, skills, tools, environment mappings, checks and workflows. Harness Library focuses on reusable, task-specific setups: from a small skill folder to an MCP workflow or a larger harness package.

Maintained by Kristofer Jussmann (@Ker102): https://github.com/Ker102

Source and review records: https://github.com/OpenRootdev/harness-library

Harneloop is optional. Setup reviews and artifact checks are scoped evidence, not universal model/platform certification.

## Can I submit just a few skills?

Yes. A useful skill folder, MCP workflow or complete harness package can qualify. Explain the task, installation and manual setup clearly.

## Do I need to use Harneloop?

No. Submissions can use any agent or framework. Built with Harneloop is an optional attribution, reviewed separately from the listing.

## What if my repository contains a whole application?

Provide a focused package or a dedicated harness branch so users can install only the harness. Do not make them download unrelated videos, caches or application assets.

## Does a listing mean it works with every model?

No. Each page states its review scope, evidence and manual prerequisites. Setup reviewed and artifacts checked are different claims; neither guarantees every platform or model.

## Is submitting free?

Yes. Browsing and submitting to the directory are free. You need a GitHub account to open a submission issue; a Harness Library account is not required.


# Smooth Motion Graphics


> A compact skill for calm, clear brand animations.


Canonical page: https://openrootdev.github.io/harness-library/harnesses/smooth-motion-graphics/


Category: Creative
Version: 1.0.0
License: MIT


Source: https://github.com/Ker102/smooth-motion-graphics
Distribution: https://github.com/Ker102/smooth-motion-graphics/releases/download/v1.0.0/smooth-motion-graphics-v1.0.0.zip
Scope: One skill, reference guide, example brief, renderer and dependency manifests. No videos or dependencies.


## Installation


### 1. Download the focused package

Download the ZIP and SHA256SUMS.txt from the release. Verify the checksum with your system's checksum tool, then extract into a folder you control.

Release and checksums: https://github.com/Ker102/smooth-motion-graphics/releases/tag/v1.0.0


### 2. Point your agent at the skill

Ask any agent to read this file. For an Agent Skills-compatible client, copy the complete skill folder to its documented skills directory. Discovery differs by client.

```

skills/smooth-motion-graphics/SKILL.md

```


### 3. Optional: render a first preview

Install Node.js 20+ and system FFmpeg for video encoding. From the skill folder, run these commands. Copy and adapt the example brief for your own brand before delivering a video.

```

npm ci
node scripts/render.cjs stills assets/brand-brief.example.json output
node scripts/render.cjs video assets/brand-brief.example.json output

```


## Manual setup

- No MCP, API key or image model is required. The instructions can be used with your existing animation tools.

- Optional reference renderer: Node.js 20+, system FFmpeg with libx264 for encoding, and your own licensed font if desired.

- Set SMOOTH_FFMPEG or SMOOTH_FONT_PATH only if using a non-default executable or local font. Inspect preview frames on the machine that will encode the video.


## Additional notes

- Reference layouts: title, 2–4 inputs into a central object, and 3–7 workflow cards with explicit directed edges. Extend them when the story needs a different layout.

- Optional Harneloop thin export is available in the release. It is a promoted snapshot for reuse; it does not include private development runs or claim a one-command import.

- The example uses fictional Luma branding. Replace the story with verified facts for your product.


## Contents

- skills: 1

- references: 1

- renderers: 1

- files: 6

- mcp_tools: 0

- tools: Node.js + @napi-rs/canvas (optional), FFmpeg (optional encoding)


## Review scope

artifact-checked; reviewed 2026-10-10.

Windows: fictional example and alternate brand frames inspected; layout/direction tests and 1080p60 encode/decode checked. Other platforms not runtime-tested.

Evidence source: https://github.com/Ker102/smooth-motion-graphics#what-gets-checked


## Built with Harneloop

Extracted, tested and promoted through an external Harneloop unit. The optional thin export contains the verified skill.

https://harneloop.kaelux.dev


# ViperMesh for Blender


> A local Blender MCP workflow for building and checking 3D scenes.


Canonical page: https://openrootdev.github.io/harness-library/harnesses/vipermesh-blender/


Category: 3D & Blender
Version: 1.3.0
License: MIT; preserve NOTICE.md attribution


Source: https://github.com/OpenRootdev/vipermesh-blender
Distribution: https://github.com/OpenRootdev/vipermesh-blender/releases/download/v1.3.0/vipermesh-blender-1.3.0.mcpb
Scope: Focused MCPB server bundle. Install the separate Blender addon and optional source skill; the full mixed repository is not needed.


## Installation


### 1. Install and enable the Blender addon

Download the addon from the release. In Blender, open Edit → Preferences → Add-ons → Install from Disk, select the downloaded Python addon file and enable ViperMesh for Blender. The maintainer currently tests Blender 5.2.

Download Blender addon: https://github.com/OpenRootdev/vipermesh-blender/releases/download/v1.3.0/vipermesh-for-blender-addon-v1.3.0.py


### 2. Connect your MCP client

Install Node.js 20+. Import the MCPB bundle into a client supporting local MCPB extensions. Keep the Blender viewport sidebar's ViperMesh Local Bridge running. Source-build/stdio configuration is documented in the repository for clients without MCPB import.

Release, checksums and alternatives: https://github.com/OpenRootdev/vipermesh-blender/releases/tag/v1.3.0


### 3. Load the agent skill and start a session

Read the skill plus its five relative reference files from the linked source folder. Copy the whole folder to your agent's documented skill location if desired. Have the agent bootstrap, inspect the existing scene and discover relevant operations before editing.

Skill and reference folder: https://github.com/OpenRootdev/vipermesh-blender/tree/main/skills/using-vipermesh-blender

```

bootstrap_vipermesh_session
get_blender_agent_context
list_blender_tools

```


## Manual setup

- A local Blender session, enabled addon, and running Local Bridge are required. This is a local workstation connector, not a hosted Blender service.

- MCP client needs stdio support; MCPB import is optional. The client should manage one persistent Node server process per agent session.

- The addon listens on loopback 127.0.0.1:9876. The connector can change scene state and exposes code execution; use it with trusted agents and approved output paths.


## Additional notes

- Existing repository and release assets are listed unchanged. Nine top-level MCP tools route to a larger Blender operation registry; nine is not the total number of Blender operations.

- The public connector and guidance are distinct from private studio capabilities and benchmark traces. Those are not included in this package.

- Retain the source license and NOTICE.md when redistributing. Consult the upstream README for current supported Blender versions.


## Contents

- skills: 1

- references: 5

- mcp_tools: 9

- tools: ViperMesh MCP server, Blender addon / Local Bridge, Node.js 20+


## Review scope

setup-reviewed; reviewed 2026-10-10.

Existing release assets, README installation and skill references reviewed. A live Blender bridge was not retested for this listing.

Evidence source: https://github.com/OpenRootdev/vipermesh-blender/releases/tag/v1.3.0