hypogenic-hypothesis-generation
Warn
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones example datasets from external GitHub repositories (
ChicagoHAI/HypoGeniC-datasetsandChicagoHAI/Hypothesis-agent-datasets) to facilitate hypothesis generation and benchmarking. - [COMMAND_EXECUTION]: The skill requires the execution of shell scripts (
setup_grobid.sh,run_grobid.sh) to install and run the GROBID service for processing literature. It also executes local Python scripts (pdf_preprocess.py) during the data preparation phase. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided tabular data and interpolates it into LLM prompt templates without sanitization or protective boundary markers.
- Ingestion points: Tabular data is loaded from JSON files defined in the task configuration (
train_data_path,val_data_path,test_data_path). - Boundary markers: The prompt templates use direct string interpolation (
${text_features_1},${label}) without delimiters or instructions to ignore malicious content within the data. - Capability inventory: The skill has the capability to perform LLM calls, write to the local file system (JSON/YAML), and execute shell commands.
- Sanitization: No validation or sanitization logic is applied to the data features before they are interpolated into the prompt strings.
- [DYNAMIC_EXECUTION]: The workflow involves the dynamic execution of environment setup and data preprocessing scripts that are bundled with or expected in the local environment.
Audit Metadata