quant-ml-trading
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [METADATA_POISONING]: The description field in SKILL.md contains an excessively long list of trading and machine learning keywords. This keyword stuffing is designed to manipulate the agent's skill selection process by matching a wide variety of user queries, potentially overriding intended selection logic.
- [INDIRECT_PROMPT_INJECTION]: The skill specifies the ingestion of market data and natural language sentiment analysis, creating a surface for indirect prompt injection attacks where instructions could be hidden in external data.
- Ingestion points: Market data feeds, OHLCV datasets, and NLP sentiment analysis sources as described in SKILL.md and referenced pipelines.
- Boundary markers: No explicit delimiters or instructions to ignore embedded data-commands are provided in the skill text.
- Capability inventory: The skill claims capabilities for persistent data storage and report generation (PDF/HTML), which could be leveraged if an injection occurs.
- Sanitization: The provided documentation does not describe any sanitization or validation processes for external market or sentiment data.
Audit Metadata