The Agent Skills Directory

[SAFE]: No security issues detected. The skill provides a robust framework for improving decision quality through controlled adversarial scrutiny.
[SAFE]: The skill incorporates human-in-the-loop checkpoints, requiring explicit user authorization before invoking subagents for cross-model review, which prevents unauthorized data transfer or cost escalation.
[SAFE]: It implements context isolation by ensuring subagents receive only the necessary artifact and contract details, which prevents confirmation bias and minimizes the data exposed to sub-tasks.
[SAFE]: The use of the task tool follows the platform's standard orchestration patterns and includes explicit instructions to avoid prohibited nested persona invocations and recursive loops.

oracle-doubt