# What AI Models Are Compatible with These Skills: A Model-Agnostic Guide to emilkowalski/skills

> Discover which AI models are compatible with emilkowalski/skills. This guide explains how to use GPT-4, Claude, Gemini, and open-source LLMs with the skills framework.

- Repository: [Emil Kowalski/skills](https://github.com/emilkowalski/skills)
- Tags: deep-dive
- Published: 2026-08-09

---

**The emilkowalski/skills repository supports any text-generation LLM, including OpenAI GPT-4, Anthropic Claude, Google Gemini, and open-source models, because it forwards skill prompts to whichever model you configure via CLI flags or environment variables.**

The `emilkowalski/skills` repository provides a framework for running automated code reviews and improvements through structured prompts defined in markdown files. Unlike tools that hard-code specific AI providers, this repository treats skills as model-agnostic definitions that work with any compatible LLM you specify during invocation.

## How the Skills Framework Handles AI Model Integration

The repository does not embed or enforce any particular large language model. According to [`README.md`](https://github.com/emilkowalski/skills/blob/main/README.md), the installation command `npx skills@latest add emilkowalski/skills` simply registers skill files locally without binding to a specific provider.

Each skill is defined as a markdown file containing metadata headers and prompt text. The CLI forwards these prompts to your chosen model at runtime, making the system inherently compatible with any text-generation AI that accepts plain-text prompts via API.

## Configuring Compatible AI Models

You specify which AI model processes your skills using either command-line arguments or environment variables. This flexibility allows you to switch between providers like OpenAI, Anthropic, or Google without modifying the skill definitions.

### Command-Line Model Selection

Pass the `--model` flag when invoking a specific skill to route the prompt to your preferred LLM:

```bash

# Run a skill using OpenAI GPT-4

skills run review-animations --model=gpt-4 \
  --input="diff of my component.tsx"

```

```bash

# Execute with Anthropic Claude 3 Sonnet

skills run improve-animations --model=claude-3-sonnet \
  --input="src/**/*.tsx"

```

### Environment Variable Configuration

Set the `SKILLS_MODEL` environment variable to configure a default model for all skill invocations in your session:

```bash
export SKILLS_MODEL=claude-3-sonnet
skills run improve-animations --input="src/**/*.tsx"

```

```bash

# Using Google Gemini via environment variable

export SKILLS_MODEL=gemini-1.5-flash
skills run animate --input="components/Button.tsx"

```

## Understanding the disable-model-invocation Flag

Some skill definitions include a `disable-model-invocation: true` flag in their metadata headers. In [`skills/review-animations/SKILL.md`](https://github.com/emilkowalski/skills/blob/main/skills/review-animations/SKILL.md), this flag appears in the frontmatter to indicate that the framework should not automatically execute the model for that specific skill.

However, this flag does **not** restrict which AI models are compatible. It merely controls whether the CLI invokes the model automatically. You can still manually run the skill with any LLM you prefer:

```bash

# Manual invocation works with any model despite the disable flag

skills run review-animations --model=mistral-large \
  --input="animation-review-input.txt"

```

## Compatible AI Model Providers and Examples

Because the repository acts as a prompt router rather than a model host, it supports any text-generation LLM accessible via API. Typical compatible models include:

- **OpenAI**: `gpt-3.5-turbo`, `gpt-4`, `gpt-4-turbo`
- **Anthropic**: `claude-2`, `claude-3-sonnet`, `claude-3-opus`
- **Google**: `gemini-1.0-pro`, `gemini-1.5-flash`, `gemini-1.5-pro`
- **Cohere**: `command-r`, `command-r-plus`
- **Mistral AI**: `mistral-large`, `mixtral-8x7b`
- **Open-source models**: Llama 2-13B, Llama 3-8B, and other hosted open-source LLMs

The only requirements are network access to the model's API endpoint and valid authentication credentials configured in your environment.

## Summary

- The **emilkowalski/skills** repository is **model-agnostic** and does not hard-code any specific LLM.
- Skills are defined in markdown files (such as [`skills/improve-animations/SKILL.md`](https://github.com/emilkowalski/skills/blob/main/skills/improve-animations/SKILL.md) and [`skills/animate/SKILL.md`](https://github.com/emilkowalski/skills/blob/main/skills/animate/SKILL.md)) containing prompts that work with any text-generation model.
- Use the `--model` CLI flag or `SKILLS_MODEL` environment variable to specify your preferred AI provider.
- The `disable-model-invocation` flag in [`skills/review-animations/SKILL.md`](https://github.com/emilkowalski/skills/blob/main/skills/review-animations/SKILL.md) controls automatic execution but does not limit model compatibility.
- Any modern LLM supporting plain-text prompts—including OpenAI GPT-4, Anthropic Claude, and Google Gemini—is compatible.

## Frequently Asked Questions

### Does the skills repository require a specific LLM?

No. The repository does not require or embed any specific language model. According to the source code in [`README.md`](https://github.com/emilkowalski/skills/blob/main/README.md), the installation process only registers skill markdown files, allowing you to route prompts to any compatible AI model you configure via the CLI.

### How do I configure OpenAI GPT-4 with skills?

You can configure GPT-4 using the `--model` flag: `skills run review-animations --model=gpt-4 --input="your-input"`. Alternatively, set the environment variable with `export SKILLS_MODEL=gpt-4` before running skills commands.

### What does the disable-model-invocation flag do?

The `disable-model-invocation: true` flag, found in files like [`skills/review-animations/SKILL.md`](https://github.com/emilkowalski/skills/blob/main/skills/review-animations/SKILL.md), instructs the CLI not to automatically invoke the AI model for that skill. This allows for manual review or preprocessing, but you can still execute the skill manually with any model using the `--model` flag.

### Can I use local open-source models with skills?

Yes. Since the framework only requires a text-generation endpoint, you can use locally hosted open-source models like Llama 3 or Mistral through local API servers (such as Ollama or vLLM), provided you configure the appropriate model identifier and endpoint in your CLI environment.