omni-model-builder
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
curlto interact with the Omni REST API for fetching and updating model definitions. - [DATA_EXFILTRATION]: The skill transmits semantic model metadata and YAML configurations to the vendor's official domain at
omniapp.co. - [INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation and modification of semantic models that include fields like
ai_context,description, andsynonyms. These fields are explicitly documented as being used to guide AI query engines (referred to as 'Blobby'), creating a potential surface for indirect prompt injection if malicious instructions are embedded within the model definitions. - Ingestion points: The skill ingests data from the Omni API (
/api/v1/models) and processes model YAML files (e.g.,.view,.topic). - Boundary markers: No specific boundary markers or delimiters are employed to separate metadata descriptions from potential instructions within the YAML fields.
- Capability inventory: The skill possesses network communication capabilities (
curl) and the ability to persist changes to the shared model. - Sanitization: The instructions do not describe any sanitization or validation of the natural language content provided for AI-facing fields.
Audit Metadata