What problem does it solve?
Long development sessions accumulate context that turns unexamined assumptions into perceived facts, leading to costly, hard-to-debug errors in production, security-sensitive systems, or irreversible operations. Catching these flaws after commit is exponentially more expensive than addressing them during decision-making.
Core Features & Use Cases
- Structured 5-step doubt cycle: Formalizes the process of surfacing claims, extracting minimal reviewable artifacts, running adversarial review, reconciling findings, and stopping at a bounded threshold to avoid endless loops.
- Fresh-context adversarial review: Eliminates author bias by using an isolated reviewer tasked exclusively with finding flaws, not validating existing reasoning.
- Optional cross-model escalation: Offers a second opinion from a different model architecture for high-stakes artifacts to catch blind spots a single model would miss.
- Common use cases: Pre-commit review of non-trivial code, validation of architectural decisions under uncertainty, verification of high-stakes system claims (e.g., thread safety, idempotence), and work in unfamiliar codebases where hidden assumptions are common.
Quick Start
Use the doubt-driven-development skill to run an adversarial review of your proposed database migration script before executing it in production.
