Red Hat AI Innovation Team — Coding Agent Plugins

Plugin marketplace for the Red Hat AI Innovation Team.

Available Plugins

Plugin	Description	Status
its-hub	Inference-time scaling for LLMs — voting, scoring, and search	Ready
sdg-hub	Synthetic data generation for LLM training and evaluation	Ready
training-hub	LLM training, continual learning, and reinforcement learning	Ready

Installation

Claude Code

Add the marketplace (one-time):

/plugin marketplace add Red-Hat-AI-Innovation-Team/plugins

Install a plugin:

/plugin install its-hub@Red-Hat-AI-Innovation-Team/plugins
/plugin install sdg-hub@Red-Hat-AI-Innovation-Team/plugins
/plugin install training-hub@Red-Hat-AI-Innovation-Team/plugins

Or browse the Discover tab to see available plugins.

Codex CLI

codex plugin marketplace add Red-Hat-AI-Innovation-Team/plugins

Then install plugins from the marketplace. For manual installation, see .codex-plugin/INSTALL.md in each plugin repo.

Using the Plugins

After installing, each plugin provides skills that auto-trigger based on your intent — just describe what you want:

• "Scale this prompt using self-consistency" → its-hub activates
• "Generate QA pairs from this document" → sdg-hub activates
• "Fine-tune this model on my data" → training-hub activates

You can also invoke skills explicitly (Claude Code: /plugin-name:skill-name, Codex: @skill-name).

Skills per Plugin

its-hub — Inference-time scaling

Skill	What it does
setup-guide	Install its_hub, configure model endpoint and algorithm
inference-scaling	Run scaling on a single prompt
batch-scaling	Run scaling on a batch of prompts from a file
inference-scaling-guide	Algorithm selection, budget tuning, reward models, troubleshooting

training-hub — LLM training

Skill	What it does
setup-guide	Install training_hub, configure algorithm, model, and data
training-guide	Run a training job using saved configuration
memory-estimation	Estimate GPU VRAM requirements before training
training-hub-guide	Algorithm selection, hyperparameters, OOM troubleshooting, loss interpretation

sdg-hub — Synthetic data generation

Skill	What it does
setup-guide	Install sdg_hub, configure LLM provider and model
data-generation	Run synthetic data generation using a flow
flow-browser	List, search, and inspect available flows
synthetic-data-generation	Approach selection, flow authoring, block reference, troubleshooting

First-Time Setup

After installing a plugin, just start using it — say "scale this prompt" or "fine-tune this model". The plugin will:

1. Detect your environment (Python, installer, GPU)
2. Install the Python library automatically if needed (asks permission first)
3. Walk you through configuration (model, endpoint, algorithm)
4. Offer smart defaults if it detects your API key (e.g., OPENAI_API_KEY)
5. Save config to a local dotfile (e.g., .its-hub/config.json)

You can also run the setup explicitly by invoking the setup-guide skill.

Try It: End-to-End Example

Here's what using its-hub looks like after installing:

You: Scale this prompt with budget 5: "What are the main causes of climate change?"

Claude: I'll run inference-time scaling on your prompt using self-consistency.

[runs its_detect.sh → library=installed, config=found]
[runs its_scale.sh --metadata "What are the main causes of climate change?"]

Selected response (5/5 agreed on the same core answer):

"The main causes of climate change are:
1. Burning fossil fuels (coal, oil, gas) for energy
2. Deforestation reducing CO2 absorption
3. Industrial processes and agriculture
4. Methane emissions from livestock and landfills"

Algorithm: self-consistency | Budget: 5 | Model: gpt-4o-mini

For first-time setup, the plugin detects your environment and offers smart defaults — if you have an OPENAI_API_KEY set, setup is a single confirmation.

Adding a Plugin

To add a new plugin to this marketplace:

1. Create a .claude-plugin/plugin.json manifest in the plugin repo
2. Place all skills in .claude/skills/ (both automation and domain knowledge)
3. Point both manifests to ./.claude/skills for discovery
4. Add an entry to .claude-plugin/marketplace.json in this repo
5. The source field should be the GitHub shorthand: Red-Hat-AI-Innovation-Team/<repo-name>

See CLAUDE.md for the full plugin architecture reference.