30-thiet-ke-master
Pass
Audited by Gen Agent Trust Hub on Jun 28, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill connects to the well-known OpenAI API at
api.openai.comto generate images when a valid API key is provided by the user. This network activity is fundamental to the skill's stated purpose and targets a trusted service provider. - [COMMAND_EXECUTION]: The skill includes logic to check for a local binary dispatcher (
$OD_BIN) to handle enterprise-level image generation. This is a legitimate integration for local infrastructure and does not involve downloading or executing untrusted remote code. - [PROMPT_INJECTION]: The skill has a standard attack surface for indirect prompt injection (Category 8) as it processes untrusted data from project files (such as
brand-guideline.md,prd.md, and campaign briefs) to construct prompts for image generation models. - Ingestion points: Reads project files (
assets/brand/,brand-guideline.md,prd.md) and user-supplied text to extract brand identity and design requirements. - Boundary markers: The instructions do not specify explicit delimiters or "ignore previous instructions" safety blocks when interpolating these brand assets into the final image generation prompt.
- Capability inventory: The skill has the capability to read local files, access the network via the OpenAI API, and invoke a local dispatcher binary if configured.
- Sanitization: There is no explicit evidence of sanitization for the extracted brand metadata before it is formatted into prompts for DALL-E 3 or MidJourney.
Audit Metadata