# Which AI Agent Frameworks Are Supported by SkillSpector?

> Discover which AI agent frameworks SkillSpector supports including Claude, Codex, Gemini, and Antigravity. Integrate seamlessly with these CLI based frameworks today.

- Repository: [NVIDIA Corporation/SkillSpector](https://github.com/NVIDIA/SkillSpector)
- Tags: getting-started
- Published: 2026-07-12

---

**SkillSpector supports four CLI-based AI agent frameworks: Claude (Anthropic), Codex (OpenAI), Gemini (Google), and Antigravity (AWS), with each framework registered in the internal provider registry at [`src/skillspector/providers/_agent_cli.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/_agent_cli.py).**

NVIDIA's SkillSpector is a security analysis tool designed to evaluate AI agent skills that execute through command-line interfaces. Understanding which AI agent frameworks are supported by SkillSpector is essential for developers who want to scan their agent implementations for vulnerabilities. The tool discovers the appropriate CLI binary automatically from skill metadata while allowing users to configure the semantic analysis backend separately.

## Supported AI Agent Frameworks

SkillSpector currently integrates with four major AI agent providers through dedicated CLI wrappers. Each framework maps to a specific provider module and binary name in the internal registry.

### Claude (Anthropic)

- **Provider Module**: `skillspector.providers.claude_cli`
- **CLI Binary**: `claude`
- **Implementation**: [`src/skillspector/providers/claude_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/claude_cli/provider.py)

This wrapper enables SkillSpector to invoke and analyze skills built for Anthropic's Claude CLI agent.

### Codex (OpenAI)

- **Provider Module**: `skillspector.providers.codex_cli`
- **CLI Binary**: `codex`
- **Implementation**: [`src/skillspector/providers/codex_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/codex_cli/provider.py)

OpenAI's Codex CLI agent is supported through this provider module, allowing analysis of Codex-based agent implementations.

### Gemini (Google)

- **Provider Module**: `skillspector.providers.gemini_cli`
- **CLI Binary**: `gemini`
- **Implementation**: [`src/skillspector/providers/gemini_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/gemini_cli/provider.py)

Google's Gemini CLI agent framework is registered under this provider for security scanning.

### Antigravity (AWS)

- **Provider Module**: `skillspector.providers.antigravity_cli`
- **CLI Binary**: `agy` (registered as "agy" in the internal registry)
- **Implementation**: [`src/skillspector/providers/antigravity_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/antigravity_cli/provider.py)

The AWS Antigravity framework uses the abbreviated binary name `agy` in the SkillSpector registry.

## Framework Registration and Detection

The registry of supported CLI agents lives in **[`src/skillspector/providers/_agent_cli.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/_agent_cli.py)**, which contains an internal `_REGISTRY` mapping framework identifiers to their respective CLI binaries. This registry explicitly references the four primary agents and serves as the source of truth for available frameworks.

When SkillSpector analyzes a skill, it uses the **`run_agent_cli`** hardened subprocess helper to safely invoke the detected CLI binary. The framework is automatically discovered from the skill's [`SKILL.md`](https://github.com/NVIDIA/SkillSpector/blob/main/SKILL.md) file, ensuring the correct binary—whether `claude`, `codex`, `gemini`, or `agy`—is launched via the secure wrapper.

## Configuring the Analysis Backend

While SkillSpector automatically detects which CLI framework to use for skill execution, you can configure the **LLM backend** for semantic analysis separately using the `SKILLSPECTOR_PROVIDER` environment variable.

To perform a basic scan without LLM analysis:

```python
import subprocess

subprocess.run(
    ["skillspector", "scan", "./my-skill/"],
    check=True,
)

```

To enable LLM-augmented analysis with a specific provider:

```python
import os
import subprocess

# Set the backend for semantic analysis

os.environ["SKILLSPECTOR_PROVIDER"] = "anthropic"  # Options include anthropic, openai, google, etc.

subprocess.run(
    ["skillspector", "scan", "./my-skill/", "--llm"],
    check=True,
)

```

The same scan commands work regardless of which CLI framework your skill uses—the provider detection operates independently from the semantic analysis backend.

## Provider Module Implementation

Each supported framework is implemented as a thin wrapper in the providers directory:

- **[`src/skillspector/providers/_agent_cli.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/_agent_cli.py)**: Core hardened subprocess helper and central registry for all supported CLI agents
- **[`src/skillspector/providers/claude_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/claude_cli/provider.py)**: Claude CLI registration
- **[`src/skillspector/providers/codex_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/codex_cli/provider.py)**: Codex CLI registration  
- **[`src/skillspector/providers/gemini_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/gemini_cli/provider.py)**: Gemini CLI registration
- **[`src/skillspector/providers/antigravity_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/antigravity_cli/provider.py)**: Antigravity CLI registration

These modules collectively define the AI agent frameworks that SkillSpector can safely evaluate.

## Summary

- SkillSpector supports **four CLI-based AI agent frameworks**: Claude (Anthropic), Codex (OpenAI), Gemini (Google), and Antigravity (AWS).
- Framework binaries (`claude`, `codex`, `gemini`, `agy`) are registered in **[`src/skillspector/providers/_agent_cli.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/_agent_cli.py)**.
- The **`run_agent_cli`** method provides a hardened subprocess wrapper for safe CLI invocation.
- Framework detection is automatic based on the skill's [`SKILL.md`](https://github.com/NVIDIA/SkillSpector/blob/main/SKILL.md) file.
- The **`SKILLSPECTOR_PROVIDER`** environment variable configures the LLM backend for semantic analysis independently from the CLI framework.

## Frequently Asked Questions

### How does SkillSpector determine which CLI binary to use for my agent?

SkillSpector automatically detects the appropriate CLI binary by reading your skill's [`SKILL.md`](https://github.com/NVIDIA/SkillSpector/blob/main/SKILL.md) file. It then references the internal `_REGISTRY` in [`src/skillspector/providers/_agent_cli.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/_agent_cli.py) to map the framework to the correct binary name—whether that's `claude`, `codex`, `gemini`, or `agy` for Antigravity.

### Can I add support for a new AI agent framework to SkillSpector?

Yes, but it requires modifying the source code. You must register the new CLI-based agent in the `_REGISTRY` dictionary within [`src/skillspector/providers/_agent_cli.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/_agent_cli.py) and create a corresponding provider module following the pattern used by the existing `claude_cli`, `codex_cli`, `gemini_cli`, or `antigravity_cli` implementations.

### What is the difference between the CLI framework and the SKILLSPECTOR_PROVIDER environment variable?

The CLI framework (detected via [`SKILL.md`](https://github.com/NVIDIA/SkillSpector/blob/main/SKILL.md)) determines which binary SkillSpector executes to run your agent skill—this is the actual AI agent being analyzed. The `SKILLSPECTOR_PROVIDER` environment variable selects which LLM backend performs the semantic security analysis of that skill's code. They operate independently: you could analyze a Claude-based skill using OpenAI's backend for semantic analysis, or vice versa.

### Is the Antigravity CLI binary name different from its framework name?

Yes, while the framework is called Antigravity (AWS), it is registered in the SkillSpector internal registry with the binary name `agy` rather than the full framework name. This abbreviation is defined in [`src/skillspector/providers/antigravity_cli/provider.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/antigravity_cli/provider.py) and referenced in the central registry at [`src/skillspector/providers/_agent_cli.py`](https://github.com/NVIDIA/SkillSpector/blob/main/src/skillspector/providers/_agent_cli.py).