Review Roulette FAQs

Question 1

Will the sub-tasks see my conversation history?

Accepted Answer

No, each review is dispatched as a fresh task with no conversation history. This ensures that every model provides an independent analysis based solely on the current code and instructions.

Question 2

What does the final output look like?

Accepted Answer

You receive a single JSON response containing a concatenated list of all suggestions and a summary section clearly labeled by which provider and model contributed each piece of feedback.

Question 3

Which models are eligible for Review Roulette?

Accepted Answer

The skill dynamically checks your environment for models with 'thinking' capabilities. It specifically excludes prohibitively expensive models like o3-pro to maintain cost-efficiency.

Question 4

Does the parent model analyze the code itself?

Accepted Answer

No, the parent model acts strictly as an orchestrator. It dispatches the full context to sub-models and merges their results without editorializing or summarizing the findings.

Question 5

How does Review Roulette ensure diverse feedback?

Accepted Answer

The skill randomly selects three different reasoning models from various providers (such as Anthropic, OpenAI, and Google) for every run, ensuring that the feedback reflects multiple independent training biases and perspectives.

Review Roulette

주요 기능

사용 사례

Review Roulette

주요 기능

사용 사례