aster-timesfm-pipeline
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads the TimesFM 2.5 forecasting model weights from Google's official repository on Hugging Face. This download is initiated by the
timesfm.TimesFM_2p5_200M_torch()call. As both Google and Hugging Face are recognized as well-known and trusted entities, this operation is considered safe and standard for the skill's machine learning purpose. - [INDIRECT_PROMPT_INJECTION]: The skill processes external market data (klines) retrieved from a DEX via the Aster MCP.
- Ingestion points: Historical kline data is ingested via
mcp__aster__get_klinesas shown inSKILL.md. - Boundary markers: None explicitly defined as prompt delimiters, though the data is handled within a structured Python context.
- Capability inventory: The pipeline performs numerical transformations and model inference; it does not involve subprocess execution, file system modifications, or arbitrary network writes based on the ingested data.
- Sanitization: The skill converts input data to numpy arrays of floats (e.g.,
float(k[4])), which effectively strips non-numeric content and prevents the model from processing unexpected natural language instructions.
Audit Metadata