together-dedicated-containers
Pass
Audited by Gen Agent Trust Hub on Apr 3, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious code or exploitable vulnerabilities were identified in the skill instructions or scripts.
- [EXTERNAL_DOWNLOADS]: References official software distributions including the together package from public registries and the sprocket SDK from Together AI's verified private index (https://pypi.together.ai/).
- [COMMAND_EXECUTION]: Provides documentation for the together beta jig CLI to perform legitimate administrative tasks such as building Docker images, managing volumes, and deploying workers.
- [CREDENTIALS_UNSAFE]: Correctly handles authentication by utilizing the TOGETHER_API_KEY environment variable and providing a workflow for managing encrypted secrets via the jig secrets command, rather than hardcoded credentials.
- [PROMPT_INJECTION]: The skill defines a data ingestion path where workers process external job payloads via the Sprocket.predict() method. This represents a potential surface for indirect prompt injection if user-implemented logic handles inputs unsafely. 1. Ingestion points: predict method in scripts/sprocket_hello_world.py. 2. Boundary markers: None present in the data schema. 3. Capability inventory: Containerized execution with network access and build-time command execution (tool.jig.image.run). 4. Sanitization: None present in templates; implementation of validation is left to the developer.
Audit Metadata