tabular-cleanup

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes user-environment code via a configurable <analysis_cmd> (defaulting to python3).
  • This is necessary for the skill's primary purpose of profiling and transforming data using the pandas library.
  • The implementation includes safety guardrails: it requests user consent for package installations via uv pip install and requires explicit user sign-off on the data contract before autonomous iterations begin.
  • [DATA_EXFILTRATION]: The skill performs local file operations restricted to the <sandbox_root> and the input <artifact>.
  • No network exfiltration patterns or unexpected external communication were detected.
  • Analysis of the data flow confirms that data is read from the source and written to a local sandbox for versioning and auditing.
  • [PROMPT_INJECTION]: The instructions contain strict operational logic and deterministic exit conditions.
  • There are no patterns of 'ignore previous instructions' or attempts to bypass safety filters.
  • The loop is designed to terminate based on data-driven checks rather than open-ended LLM heuristics.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 04:41 PM
Security Audit — agent-trust-hub — tabular-cleanup