cumcm-master
Warn
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill instructs the agent to autonomously generate Python scripts within a workspace (e.g.,
01_data_eda.py,02_problem1_model.py) and execute them using the system shell to perform data analysis, modeling, and visualization tasks. - [COMMAND_EXECUTION]: The skill uses various shell commands for environment setup (
bootstrap_runtime.py), workspace management (setup_workspace.py), state tracking (agent_memory_manager.py), and LaTeX paper compilation (xelatex). - [INDIRECT_PROMPT_INJECTION]: The agent ingests untrusted data from external sources which directly influences its reasoning and the code it generates for execution.
- Ingestion points: Processes user-provided PDF or text files for problem statements and fetches content from the web via
WebSearchandWebFetchtools. - Boundary markers: Absent; there are no specified delimiters or instructions to prevent the agent from following directives embedded within the ingested problem statements or research data.
- Capability inventory: Extensive capabilities including writing files, executing shell commands, and running dynamically generated Python scripts.
- Sanitization: Absent; no methods are defined to sanitize or validate the content extracted from external documents or web search results before it is used in the prompt context.
Audit Metadata