What problem does it solve? Confident AI outputs are not always correct, and long sessions quietly turn assumptions into accepted facts. This Skill catches wrong decisions early by forcing every non-trivial decision through an adversarial fresh-context review while course-correction is still cheap. ## Core Features & Use Cases - Five-step doubt cycle: CLAIM the decision, EXTRACT the smallest reviewable artifact plus contract, DOUBT it with an adversarial reviewer, RECONCILE findings by precedence, and STOP within a bounded three-cycle loop. - Cross-model second opinions: Optionally escalate reviews to external CLI models like Gemini or Codex with read-only sandboxing, stdin piping, and explicit per-invocation user authorization. - Scope guardrails: Applies only to non-trivial decisions such as branching logic, cross-boundary changes, and irreversible operations, explicitly excluding mechanical edits and trivial changes. - Use Case: Before committing a new caching layer claimed to be thread-safe, extract the diff and its contract, hand them to a fresh-context reviewer with an adversarial prompt, classify the findings, and fix real issues before merging. ## Quick Start Ask the AI to apply doubt-driven development to review the thread-safety of the caching change you are about to commit.
