huawei-cloud-cloudrobo-model-workflow
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions and reference documents recommend using the
cloudroboCLI for all orchestration tasks. To avoid shell-specific parsing issues with JSON parameters, the skill utilizes Python'ssubprocessmodule to call the CLI. This is a standard integration method for cloud platform tooling. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes external data including user-provided task descriptions and metadata from training datasets. However, the risk is mitigated as the skill operates within a controlled cloud environment using structured CLI commands.
- Ingestion points: Data enters through user-defined task descriptions, robot type specifications, and dataset metadata files such as
meta/info.jsonandREADME.md. - Boundary markers: The skill relies on standard CLI parameterization and JSON structure to delineate data.
- Capability inventory: The agent can create training tasks, deploy inference services, and dispatch robot evaluation tasks via the
cloudroboCLI usingsubprocess.run. - Sanitization: The skill emphasizes adhering to specific platform JSON schemas and compact formatting for validation.
Audit Metadata