ai-vector-brain
Fail
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: CRITICALINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external document corpora through ingestion scripts (inventory_corpus.py, chunk_corpus_files.py), creating an attack surface for instructions embedded in ingested data.
- Ingestion points: scripts/inventory_corpus.py, scripts/prepare_documents.py, and scripts/chunk_corpus_files.py read content from local directories.
- Boundary markers: SKILL.md includes explicit instructions to treat retrieved content as untrusted.
- Capability inventory: The skill uses psycopg for SQL operations and subprocess in tests.
- Sanitization: The scripts do not appear to perform explicit content sanitization on the document content before it is stored or processed.
- [EXTERNAL_DOWNLOADS]: The documentation (references/deferred-extensions.md) contains a reference to trulens.org, a well-known service for RAG evaluation. This domain has been noted by automated scanners for inclusion in certain blocklists.
- [COMMAND_EXECUTION]: Development and testing utilities (scripts/test_build_vector_hub.py, scripts/test_check_brain_manifest.py) utilize subprocess.run() to execute local Python and Shell scripts for regression verification purposes.
Recommendations
- Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata