lambda-labs-gpu-cloud
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTIONPERSISTENCEPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for the agent to ingest and execute code from untrusted external sources, such as cloning git repositories and installing dependencies from requirements.txt files. This creates a surface where malicious content could influence agent behavior.
- Ingestion points: SKILL.md (instructions for git clone and pip install -r requirements.txt).
- Boundary markers: Absent.
- Capability inventory: Subprocess execution, file system writes, and Python/pip execution.
- Sanitization: Not present.
- [REMOTE_CODE_EXECUTION]: The documentation includes instructions to download and execute an installation script from the official Lambda Labs domain (lambdalabs.com) for setup and troubleshooting.
- [PERSISTENCE]: Provides instructions to modify shell profile files (~/.bashrc) and SSH authorized_keys to maintain persistent environment variables and access configurations.
- [PRIVILEGE_ESCALATION]: Utilizes sudo for administrative tasks like rebooting the instance or installing system-level packages.
- [COMMAND_EXECUTION]: Employs shell commands extensively for instance management, connectivity, and executing machine learning workloads.
- [EXTERNAL_DOWNLOADS]: Fetches software, models, and configurations from external sources including GitHub and the Lambda Labs official domain.
Audit Metadata