doubt-driven-development

Adversarial code review validates decisions by focusing on disproof rather than approval. It subjects non-trivial code or design choices to rigorous review, using a fresh-context reviewer to catch blind spots and improve correctness early.

To prevent context-induced blind spots during code review, invoke a fresh-context reviewer. This approach systematically evaluates non-trivial decisions without prior bias, ensuring issues are caught early and system reliability is improved.

Yes, this adversarial review process requires Claude Code and role-based agents located in the .shotgun/agents/ directory. These dependencies are necessary to support cross-model review and interactive verification sessions.

To review caching layer thread-safety, run the skill with the specific decision artifact. A structured checklist process systematically validates the non-trivial design choices through an interactive session using a fresh-context reviewer.

Yes, you can request cross-model reviews for further verification. During the interactive session, the skill invokes role-based agents from .shotgun/agents/ to provide additional layers of decision validation and correctness assurance.

Adversarial review should not be used for trivial code changes that lack significant design decisions. It is specifically designed to rigorously validate non-trivial decisions where early issue detection improves overall system reliability.