byted-emr-skills
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]:
scripts/on_serverless/emr_serverless_submit_cli.pyuses the__import__function to dynamically load task modules from theserverlessSDK based on hardcoded mapping lists. This enables the skill to handle various job types like Spark and Ray through dynamic dispatch. - [EXTERNAL_DOWNLOADS]: The skill installs a local Python wheel asset (
assets/libs/python_serverless-1.0.3.4.1-py3-none-any.whl) usingpip installvia thescripts/bin/install_serverless_sdk.shscript. This package is a ByteDance-developed SDK required for EMR Serverless operations. - [COMMAND_EXECUTION]: The skill provides CLI-style Python scripts that act as wrappers for Volcano Engine OpenAPI endpoints. These allow the agent to manage cluster nodes, submit big data jobs, and update service configurations.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data from EMR diagnostic tools and logs, which represents a potential attack surface.
- Ingestion points: Diagnostic reports and session histories fetched by
scripts/emr_agent/expert.pyandscripts/emr_agent/emr_agent_manager.py. - Boundary markers: None identified; external data is integrated into the output stream without specific delimiters.
- Capability inventory: Cluster resource management, job execution (Spark/Ray), and file management within the
./tmpdirectory. - Sanitization: Content from external APIs is displayed directly; no specific sanitization for malicious prompts in logs was observed.
Audit Metadata