doubt-driven-development

Adversarial decision review catches hidden assumptions by spawning an unbiased reviewer who receives only the artifact and contract to disprove your logic, identifying edge cases and hidden coupling that accumulated context hides. It classifies all findings as contract misreads, actionable issues, valid trade-offs, or noise before stopping.

Adversarial fresh-context review validates irreversible operations by subjecting data migrations and public API updates to a bounded 3-cycle review loop, catching flaws before finalization. This prevents costly production bugs and security vulnerabilities that are expensive to fix after deployment.

Adversarial code review for security-sensitive logic works by requiring reviewers to receive only the artifact and contract, not the author's reasoning, to avoid bias. It enforces a bounded 3-cycle loop with optional cross-model escalation via external CLIs to catch blind spots shared by a single model.

You need adversarial review for architectural decisions when verification cost is lower than debugging cost, especially for complex code changes, security-sensitive logic, irreversible operations including data migrations and public API updates, and work in unfamiliar codebases where unstated assumptions lead to costly production bugs.

Yes, optional cross-model escalation offers multi-model review via external CLIs or the consensus skill for high-stakes artifacts, catching blind spots shared by a single model. It includes strict safety guardrails for external tool execution during the adversarial review process.

The bounded 3-cycle review loop limits rework by enforcing a maximum of three cycles with clear stop conditions including trivial findings, user override, or reaching the cycle limit. This prevents endless rework while ensuring thorough scrutiny of non-trivial development decisions.