ml-data-pipeline-architecture

Pass

Audited by Gen Agent Trust Hub on Apr 4, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill documentation includes an explicit instruction on how to bypass a platform-level constraint ('PreToolUse hook') that enforces a preference for the Polars library. This allows overriding the restriction by adding a specific comment to the file top.
  • Evidence: 'To use Pandas, add # polars-exception: at file top.' in SKILL.md.
  • [PROMPT_INJECTION]: Indirect prompt injection surface analysis.
  • Ingestion points: Data is ingested from ClickHouse and Parquet files in the provided Python code snippets in SKILL.md.
  • Boundary markers: The code templates lack markers or instructions to delimit external data from the execution context.
  • Capability inventory: The skill is configured with filesystem tools including Read, Grep, and Glob.
  • Sanitization: No data validation or sanitization logic is present in the provided ingestion patterns.
  • [SAFE]: External documentation references target well-known technology projects and services.
  • Evidence: Links to docs.pola.rs, arrow.apache.org, clickhouse.com, pytorch.org, and pypi.org in SKILL.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 4, 2026, 09:51 AM