Back to skill

Security audit

freegen-image-generation

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a straightforward image-generation guide, but it explicitly tells users how to route restricted prompts through an unfiltered model.

Review this skill before installing if you need content-safety guarantees. Do not use it to bypass provider or platform safety rules, and avoid sending sensitive, personal, regulated, or confidential prompts or images unless you are comfortable with them being processed by the local freegen service and Dezgo.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

other

Warning
Location
SKILL.md:193
Finding

Explicit Content-Filter Circumvention Guidance

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 193–198
Vulnerability Type: Content Safety Filter Circumvention
Risk Level: Medium

Complete Snippet:

markdown
**Dezgo Content Policy (Stand 28.07.2026):**

| API | Content-Filter |
|---|---|
| Flux TXT2IMG | Kein Filter – alle Prompts erlaubt |
| SDXL/SD1.5/SDXL2 TXT2IMG | Standard Dezgo-Filter |
| **Image-Tools** (I2I, Inpainting, RemoveBG, Upscale, Edit, ControlNet) | Dezgo blockt bestimmte Prompts (HTTP 500) |

**Workaround für restriktive Prompts:**
- Flux TXT2IMG mit direktem Prompt (kein Filter)

Technical Analysis

The Skill identifies Flux TXT2IMG as an unfiltered route where all prompts are accepted and explicitly recommends it as a workaround for prompts restricted by other Dezgo endpoints. This creates a documented method for circumventing upstream content controls.

The same behavior is reinforced at lines 47 and 207, where Flux is advertised as “uncensored.” This issue does not qualify as skill instruction hijacking because the Skill does not attempt to alter the Agent's governing instructions, role, or safety hierarchy. It is instead a content-safety circumvention issue outside the listed T01–T09 classifications.

Attack Path

  1. A user submits an image-generation prompt prohibited by a filtered Dezgo model.
  2. The filtered endpoint rejects or blocks the prompt.
  3. The Agent consults the Skill and follows its documented workaround.
  4. The Agent resubmits the restricted prompt directly through the unfiltered Flux TXT2IMG category.
  5. The service may generate content that the filtered endpoint was intended to prevent.

Impact Assessment

Exploitation does not grant operating-system privileges, code execution, persistent access, or access to credentials. Its scope is the image-generation service and the content it can produce. A user may bypass model-level filtering and potentially generate restricted imagery, ...[truncated 77 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the recommendation to route restricted prompts through an unfiltered model.
  2. Replace “all prompts allowed” and “uncensored” language with a clear acceptable-use policy.
  3. Require policy validation before requests are submitted, regardless of the selected model or upstream filtering behavior.
  4. Treat an upstream content-policy rejection as final rather than retrying the prompt through a less restrictive endpoint.
  5. Apply consistent server-side moderation to every generation category, including Flux.
  6. Record rejected requests and model-routing decisions for abuse monitoring without retaining unnecessary sensitive prompt or image data.
  7. Document that the Agent's governing safety requirements remain applicable even when an upstream model does not enforce equivalent filters.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill states that a local container wraps the Dezgo API, which means user prompts and potentially image data are relayed to an external third-party service, but it does not clearly warn users about that transmission. This creates privacy and data-handling risk because users may submit sensitive prompts or images without informed consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The documented image-editing and inpainting flows encourage uploading user-supplied images for modification but provide no warning about privacy, retention, or the risks of submitting sensitive images. Because these operations commonly involve personal photos, omission of a caution materially increases the chance of exposing personal or confidential visual data.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The documentation explicitly advertises an 'uncensored' generation path and frames it as a way around provider filtering. That materially increases abuse potential by enabling prohibited or harmful image generation and normalizing policy evasion within the skill instructions.

Content

No source excerpt is available for this finding.

Ssd 4

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documentation gives a concrete workaround for restrictive prompts by directing users to use Flux because it has 'kein Filter.' This is effectively guidance for bypassing content restrictions and makes harmful misuse easier by embedding an evasion path into normal usage instructions.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 224)May include surrounding context.

javascript
// Bild generieren
const res = await fetch('http://freegen:3000/generate', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

Most of the instructional content is presented in German, while the skill name and API-oriented context are otherwise broadly applicable. Under the policy rule, forcing a specific language without user opt-in or a documented regional justification is a natural-language policy concern.

Content

No source excerpt is available for this finding.