light-diagram
Fail
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: CRITICALDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DATA_EXFILTRATION]: The automated scanner detected a blacklisted malicious URL
https://sub.zzshu.cc/v1/images/generationswithin the skill's evaluation records (evaluations/2026-11-02-custom-service-test/attempt-01/generation.json). The scriptlight-diagram-prompt/scripts/image_service.pyis designed to send user-configured API keys to such endpoints via theAuthorization: Bearerheader. - [COMMAND_EXECUTION]: The skill utilizes
win32com.clientin several scripts (e.g.,light-visio-diagram/scripts/render_scene.py,light-powerpoint-diagram/scripts/draw_visible.py,light-illustrator-diagram/scripts/illustrator.py) to automate desktop applications. This capability allows the agent to perform arbitrary actions within the user's local installations of Visio, PowerPoint, and Adobe Illustrator. - [DYNAMIC_EXECUTION]: The script
light-illustrator-diagram/scripts/illustrator.pydynamically assembles and executes ExtendScript code within Adobe Illustrator using theapp.DoJavaScript(code)method. This represents a dynamic code generation and execution vector. - [INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect prompt injection.
- Ingestion points: It is designed to read and parse research papers, methods sections, and user-supplied concepts (documented in
light-diagram-prompt/SKILL.md). - Boundary markers: While it uses structured data like
scene.jsonand internal IDs, it relies on the agent to interpret potentially untrusted external text. - Capability inventory: The skill possesses significant capabilities, including network access for image generation and filesystem manipulation via office automation tools.
- Sanitization: The
image_service.pyscript performs basic JSON encoding but does not sanitize the conceptual content of the prompt derived from external sources. - [EXTERNAL_DOWNLOADS]: The
image_service.pyscript performs network operations usingurllib.requestto download image assets from remote URLs returned by generation services, potentially exposing the agent to malicious payloads or tracking.
Recommendations
- AI detected serious security threats
- Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata