What problem does it solve?
Doubt driven development prevents AI coding sessions from shipping optimistic claims that cannot be proven, by forcing a bounded, adversarial review grounded in concrete evidence before release.
Core Features & Use Cases
- Evidence-backed skeptical review: Converts delivery claims into falsifiable statements, then challenges them using memory, code search, CodeGraph state, receipts, logs, receipts, and verification outputs.
- Bounded adversarial loop with clear STOP: Runs up to three structured cycles (DOUBT → RECONCILE → targeted verification) and ends with an explicit verdict rather than rubber-stamping.
- Release/security/workflow gating: Escalates to blockers for missing proof, authorization gaps, receipt gaps, security/privacy risks, contract breaks, unhandled edge states, or degraded evidence, and records residual risk when allowed.
Quick Start
Use doubt-driven-development after implementation to run a bounded skeptical review for your release claim by asking the AI tool to execute the doubt-review for the current work item and its production claim.