data-collection-sim
Warn
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The script scripts/production_capture_qa.py uses numpy.load with allow_pickle=True to read bounding box metadata. This constitutes unsafe deserialization and could allow arbitrary code execution if the files in the output directory are maliciously modified.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data through configuration files and CLI arguments, creating a surface for indirect injection.
- Ingestion points: YAML configuration files via the --config argument in scripts/production_capture_qa.py and scripts/warehouse_sdg.py, and positional arguments in scripts/validate_sdg_output.sh.
- Boundary markers: No explicit delimiters or instructions are provided to the agent to treat external inputs as untrusted data.
- Capability inventory: The skill allows file writing through Replicator backends and possesses local shell execution capabilities for validation.
- Sanitization: Configuration files are parsed using yaml.safe_load(), which is a positive safety measure, but CLI arguments in the shell script lack rigorous validation.
Audit Metadata