ml4t-research-operator
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The content is purely informational and serves as a blueprint for implementing a 'Research Operator' pattern. It does not contain any executable scripts, hidden payloads, or instructions to bypass safety protocols.
- [COMMAND_EXECUTION]: While the skill mentions tool capabilities like
run_bashandedit_file, these are presented within a conceptual framework that explicitly advocates for sandboxing, timeouts, and restricted access to prevent production risks. - [INDIRECT_PROMPT_INJECTION]: The skill identifies the surface area where indirect prompt injection could occur (e.g., reading artifacts or registry data) and recommends using typed tools and boundary checks as a defense-in-depth strategy.
- [DATA_EXFILTRATION]: No exfiltration patterns were found. The 'CORRECT' implementation example emphasizes the use of
readonly_sql_queryand sandboxed file reads to prevent unauthorized data exposure.
Audit Metadata