The Agent Skills Directory

[SAFE]: The skill facilitates the use of NVIDIA cuOpt for linear programming and routing optimization. It uses standard configuration practices for connecting to remote services and internal host endpoints (host.openshell.internal).
[COMMAND_EXECUTION]: The skill instructs the agent to use bash -lc to run local diagnostic scripts and user-defined solver scripts. These operations are necessary for the skill's functionality and are restricted to the sandbox environment.
[PROMPT_INJECTION]: The skill processes user-provided CSV data to formulate models, which is an intended functionality but creates a surface for indirect prompt injection.
- Ingestion points: Tabular files uploaded to the workspace as described in references/activation.md.
- Boundary markers: Not explicitly defined in this skill's reference files.
- Capability inventory: Script execution for interfacing with the remote gRPC/REST solver.
- Sanitization: Not specified.

cuopt-sandbox