golden-dataset-management
Fail
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: CRITICALDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: Contains references to URLs flagged as malicious by automated scanners (https://project.dev/placeholder/123 and https://project.dev/placeholder/doc-123). While these strings are explicitly documented as examples of "wrong" or invalid placeholder data to be filtered out by the skill's logic, their presence in the skill source is identified due to their blacklisted status.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest external data (URLs and document content) for storage and retrieval, presenting an attack surface for indirect injection. \n
- Ingestion points: Data is loaded from external sources or JSON backups into the Analysis and Chunk database models (SKILL.md, templates/backup-script.py). \n
- Boundary markers: The implementation lacks explicit delimiters or instructions to the model to ignore potential injection patterns within the dataset content. \n
- Capability inventory: The skill possesses capabilities for database operations (SQLAlchemy), file system writes (JSON backup creation), and network interactions (OpenAI API calls for embedding generation). \n
- Sanitization: While the skill validates URLs against placeholders, it does not describe sanitization or safety filtering for the actual document content before processing it for retrieval.
Recommendations
- Contains 2 malicious URL(s) - DO NOT USE
Audit Metadata