ai-vector-brain
Fail
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: CRITICALINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of Retrieval-Augmented Generation (RAG) systems, which by nature ingest untrusted data from external repositories, documentation, or policy files to serve as context for an AI agent.
- Ingestion points: Untrusted data enters the context through
scripts/inventory_corpus.py,scripts/chunk_markdown.py, andscripts/chunk_corpus_files.pywhich process external files into database records. - Boundary markers: The
references/agent-tool-contract.mdspecifies that the agent should treat retrieved content as evidence rather than instructions, though enforcement depends on the downstream generator model. - Capability inventory: The skill scripts possess capabilities for file system reads, database writes/reads (
scripts/embed_and_load.py), and network communication with embedding providers. - Sanitization: The skill relies on external embedding models and the agent's internal guardrails for content safety, as scripts do not perform deep content sanitization of ingested documents.
- [EXTERNAL_DOWNLOADS]: The skill interfaces with several well-known and reputable AI service providers to generate embeddings and perform reranking.
- Communication occurs with official endpoints for OpenAI, Voyage AI, Cohere, Google Gemini, and Amazon Bedrock.
- References are provided for TruLens, a well-known evaluation framework for large language model applications.
- [COMMAND_EXECUTION]: The skill utilizes system commands for database operations and internal script orchestration.
scripts/verify_postgres_memory.pyexecutes thepsqlutility to perform behavioral verification of SQL assets. This script implements strict regex validation for SQL identifiers ([a-z][a-z0-9_]*) and handles connection strings securely to prevent shell injection or credential exposure.scripts/build_vector_hub.shuses shell commands to coordinate the execution of multiple Python scripts during the artifact generation process.- Several Python test files (
scripts/test_build_vector_hub.py,scripts/test_check_brain_manifest.py, etc.) utilizesubprocess.runto validate the skill's CLI tools.
Recommendations
- Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata