Atlas Cloud — AI Image, Video & LLM Generation API

• Outcome: Pass
• Latest audit: May 29, 2026 · 8:16 PM UTC
• Version: 1.0.6

Overview

This is a disclosed cloud AI API helper that sends prompts, media, and chat requests to Atlas Cloud, with no hidden persistence or destructive behavior found.

Install only if you are comfortable sending prompts, media URLs, uploaded files, and chat messages to Atlas Cloud for processing and having generated outputs hosted on its CDN. Use revocable API keys, monitor billing, avoid confidential or regulated data unless your policy permits it, and be aware that the image script can also use a Gemini API key if you explicitly use the Google provider.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (20)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 91% confidence
• Finding: The skill declares environment-variable requirements and clearly instructs users to make authenticated network requests, yet the static finding indicates those capabilities are not properly declared as permissions. That creates a transparency and governance problem: the skill can access secrets and transmit data externally without clear permission scoping, increasing the chance of unreviewed secret use and unexpected outbound data flow.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 87% confidence
• Finding: The file markets broad functionality that is not actually implemented, including unified multi-provider support, LLM access, and pricing comparison behavior. Security-wise, this is dangerous because users and orchestration systems may route unrelated requests to this skill under false assumptions, causing unintended external transmission of prompts, media, or credentials to a third-party service.

Vague Triggers

Medium

• Confidence: 89% confidence
• Finding: The activation language is extremely broad and is designed to match many generic AI-related requests, not just narrow Atlas Cloud operations. In context, that broad matching increases the probability that sensitive user prompts, files, or development tasks are routed to this third-party API skill unnecessarily, expanding data-exposure risk.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The file instructs users to send prompts and an API key to a third-party remote service but does not warn that prompts, system messages, and possibly sensitive user data will leave the local environment. In an agent skill context, this omission can lead users to unknowingly transmit confidential data to Atlas Cloud.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The file instructs users to send prompts and optional image URLs to a third-party API but provides no warning that these inputs may contain sensitive business, personal, or copyrighted data. In a skill designed to be copied into real projects, this omission can cause unintended disclosure because users may assume sample code is safe for arbitrary inputs.

External Transmission

Medium

• Category: Data Exfiltration

Content

```bash
# Submit task
curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "bytedance/seedream-v5.0-lite", "prompt": "A cherry blossom garden", "size": "2048*2048"}'

• Confidence: 97% confidence
• Finding: This example explicitly sends user prompts and an authorization bearer token to an external Atlas Cloud endpoint. In this skill's context, external transmission is expected, but it is still a real security concern because prompts and potentially sensitive media are disclosed to a third party and tied to an account credential.

External Transmission

Medium

• Category: Data Exfiltration

Content

| Type | Endpoint | Method |
|------|----------|--------|
| **Image Generation** | `https://api.atlascloud.ai/api/v1/model/generateImage` | POST |
| **Video Generation** | `https://api.atlascloud.ai/api/v1/model/generateVideo` | POST |
| **Poll Result** | `https://api.atlascloud.ai/api/v1/model/prediction/{id}` | GET |
| **LLM Chat** | `https://api.atlascloud.ai/v1/chat/completions` | POST |

• Confidence: 95% confidence
• Finding: The documented API architecture defines multiple outbound endpoints for image, video, polling, and chat operations. This is a real external data-transfer surface because using the skill necessarily transmits prompts, media, and authentication material to Atlas Cloud services.

External Transmission

Medium

• Category: Data Exfiltration

Content

| Type | Endpoint | Method |
|------|----------|--------|
| **Image Generation** | `https://api.atlascloud.ai/api/v1/model/generateImage` | POST |
| **Video Generation** | `https://api.atlascloud.ai/api/v1/model/generateVideo` | POST |
| **Poll Result** | `https://api.atlascloud.ai/api/v1/model/prediction/{id}` | GET |
| **LLM Chat** | `https://api.atlascloud.ai/v1/chat/completions` | POST |
| **Model List** | `https://console.atlascloud.ai/api/v1/models` | GET (no auth) |

• Confidence: 95% confidence
• Finding: The listed video-generation endpoint is another explicit external transmission path. Given the skill's purpose, this is expected rather than malicious, but it still presents confidentiality and policy risk because user-supplied prompts and possibly derived media are processed by a remote provider.

External Transmission

Medium

• Category: Data Exfiltration

Content

|------|----------|--------|
| **Image Generation** | `https://api.atlascloud.ai/api/v1/model/generateImage` | POST |
| **Video Generation** | `https://api.atlascloud.ai/api/v1/model/generateVideo` | POST |
| **Poll Result** | `https://api.atlascloud.ai/api/v1/model/prediction/{id}` | GET |
| **LLM Chat** | `https://api.atlascloud.ai/v1/chat/completions` | POST |
| **Model List** | `https://console.atlascloud.ai/api/v1/models` | GET (no auth) |

• Confidence: 94% confidence
• Finding: Polling the prediction endpoint continues the external interaction and may expose generated asset identifiers and result metadata to the remote service. In aggregate with submission endpoints, this confirms persistent third-party processing rather than a local-only helper.

External Transmission

Medium

• Category: Data Exfiltration

Content

| **Image Generation** | `https://api.atlascloud.ai/api/v1/model/generateImage` | POST |
| **Video Generation** | `https://api.atlascloud.ai/api/v1/model/generateVideo` | POST |
| **Poll Result** | `https://api.atlascloud.ai/api/v1/model/prediction/{id}` | GET |
| **LLM Chat** | `https://api.atlascloud.ai/v1/chat/completions` | POST |
| **Model List** | `https://console.atlascloud.ai/api/v1/models` | GET (no auth) |

All requests (except Model List) require:

• Confidence: 94% confidence
• Finding: The chat-completions endpoint is an explicit path for transmitting conversational content to an external LLM provider. This is especially sensitive because chat payloads often contain secrets, source code, customer data, or internal context that users may not expect to leave their environment.

External Transmission

Medium

• Category: Data Exfiltration

Content

```bash
# Submit task
curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "bytedance/seedream-v5.0-lite", "prompt": "A cherry blossom garden", "size": "2048*2048"}'

• Confidence: 97% confidence
• Finding: This example explicitly sends user prompts and an authorization bearer token to an external Atlas Cloud endpoint. In this skill's context, external transmission is expected, but it is still a real security concern because prompts and potentially sensitive media are disclosed to a third party and tied to an account credential.

External Transmission

Medium

• Category: Data Exfiltration

Content

# Returns {"code": 200, "data": {"id": "prediction_xxx"}}

# Poll result (every 3 seconds)
curl -s "https://api.atlascloud.ai/api/v1/model/prediction/{prediction_id}" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY"
# When done: {"data": {"status": "completed", "outputs": ["https://cdn..."]}}
```

• Confidence: 96% confidence
• Finding: The polling example shows continued authenticated requests to retrieve results, confirming that generated outputs and job metadata are hosted and accessed externally. That raises confidentiality and lifecycle concerns, especially because the skill notes outputs are hosted on Atlas Cloud CDN.

External Transmission

Medium

• Category: Data Exfiltration

Content

### Video Generation

```bash
curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateVideo" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kwaivgi/kling-v3.0-std/text-to-video", "prompt": "A rocket launching", "duration": 5, "aspect_ratio": "16:9"}'

• Confidence: 96% confidence
• Finding: The video-generation example transmits prompts to a remote service for asynchronous processing, which can involve longer-lived storage and broader metadata collection than a simple synchronous call. In this context, the risk is real but intrinsic to the advertised service rather than evidence of malice.

External Transmission

Medium

• Category: Data Exfiltration

Content

client = OpenAI(
    api_key="your-atlascloud-api-key",
    base_url="https://api.atlascloud.ai/v1",
)
response = client.chat.completions.create(
    model="qwen/qwen3.5-397b-a17b",

• Confidence: 96% confidence
• Finding: The OpenAI-compatible client example directs chat traffic to Atlas Cloud via a custom base URL, making external transmission easy to integrate into applications. That convenience increases the chance developers unknowingly route sensitive prompts or code to a third-party endpoint under an OpenAI-like interface.

Context Leakage

High

• Category: Data Exfiltration

Content

| `atlas_get_model_info` | Get detailed API docs and schema for a model |
| `atlas_generate_image` | Submit image generation task |
| `atlas_generate_video` | Submit video generation task |
| `atlas_chat` | Send chat completion request |
| `atlas_get_prediction` | Check generation status and get results |
| `atlas_quick_generate` | One-step: auto-find model + generate |

• Confidence: 92% confidence
• Finding: The MCP tool `atlas_chat` is described as sending chat completion requests, which implies forwarding conversation content to an external service. In skill contexts, chat tools are particularly sensitive because they may be invoked with rich conversational context, leading to leakage of user data, internal instructions, source code, or other secrets to Atlas Cloud if boundaries are not strictly enforced.

External Transmission

Medium

• Category: Data Exfiltration

Content

"""
    # Step 1: Submit generation task
    payload = {"model": model, "prompt": prompt, **kwargs}
    resp = requests.post(f"{BASE_URL}/model/generateVideo", json=payload, headers=HEADERS, timeout=50)
    resp.raise_for_status()
    data = resp.json()

• Confidence: 94% confidence
• Finding: This code transmits user-supplied prompt data and arbitrary extra parameters to an external service. That is expected for an API client, but it is still a real data-exfiltration risk if callers pass sensitive text, internal URLs, or private image references without informed consent or validation.

External Transmission

Medium

• Category: Data Exfiltration

Content

---

## cURL

### Text-to-Video

• Confidence: 90% confidence
• Finding: The cURL example demonstrates direct submission of prompt data to an external endpoint using a bearer token. As documentation, this is not covertly malicious, but it normalizes copying commands that may expose sensitive prompts and media references to a remote provider without any warning about privacy implications.

External Transmission

Medium

• Category: Data Exfiltration

Content

### Image-to-Video

```bash
PREDICTION_ID=$(curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateVideo" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

• Confidence: 90% confidence
• Finding: This line includes the actual outbound cURL request for image-to-video generation, transmitting prompt content and an image URL to Atlas Cloud. The risk is heightened because image URLs may point to sensitive or access-controlled resources that users do not realize will be fetched by a third party.

External Transmission

Medium

• Category: Data Exfiltration

Content

### Text-to-Video

```bash
PREDICTION_ID=$(curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateVideo" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

• Confidence: 89% confidence
• Finding: This cURL submission example performs an actual outbound request carrying user prompt content to a third-party API. In context this is legitimate functionality, but it still presents a real confidentiality risk if users paste sensitive data into the sample command without warnings.

External Transmission

Medium

• Category: Data Exfiltration

Content

### Image-to-Video

```bash
PREDICTION_ID=$(curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateVideo" \
  -H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

• Confidence: 90% confidence
• Finding: This line includes the actual outbound cURL request for image-to-video generation, transmitting prompt content and an image URL to Atlas Cloud. The risk is heightened because image URLs may point to sensitive or access-controlled resources that users do not realize will be fetched by a third party.

VirusTotal

57/57 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.