remote-compute-modal
Pass
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads scientific software, model weights, and code from various sources including GitHub, AWS S3, and academic mirrors (e.g., ipd.uw.edu, boltz.bio). These downloads are integrated into the environment setup process for building Modal images. Many of these resources are verified using hardcoded SHA256 hashes to ensure integrity.
- [REMOTE_CODE_EXECUTION]: Facilitates the execution of user-supplied commands and scientific pipelines in remote Modal sandboxes. This behavior is the primary intended purpose of the skill and is governed by user approval cards and platform-level security controls.
- [COMMAND_EXECUTION]: The provider implementation and environment scripts use shell commands (e.g., git, wget, aria2c, pip, sed) to manage remote container lifecycles, install packages, and configure runtime environments.
- [DATA_EXFILTRATION]: While the skill connects to Modal's orchestration API, it includes specific mechanisms to enforce network confinement. It supports domain allowlisting and total network blocking for remote jobs to prevent unauthorized data exfiltration from the compute environment.
- [SAFE]: The skill implements several security-enhancing patches to the Modal SDK, such as enforcing timeout caps, preventing unauthorized GPU usage in setup kernels, and sanitizing ownership tags to ensure sandbox tenancy boundaries.
Audit Metadata