invisible-error-detect

To detect hidden runtime and logic errors in code changes, combine forward code-driven analysis with backward failure-driven reasoning. This dual-agent approach identifies logic gaps, runtime triggers, and scenario-specific vulnerabilities that standard code reviews typically miss before handoff verification completes.

The best way to run invisible error detection on security-sensitive modules and migrations is executing a dual-agent parallel analysis during handoff verification. This process merges independent forward and backward findings, applies noise gating, and generates a structured report with PASS, WARN, or BLOCK verdicts for prioritized fixes.

Invisible error detection works automatically based on git diff size and sensitive paths. It triggers conditional execution for medium-to-large code changes, while also offering a --deep-check override for forced runs and a --lite mode that simulates both analytical perspectives without spawning additional agents.

Reduce false positives when finding missed failure modes by applying structured deduplication and a 2-of-4 noise gate. The system normalizes findings, cross-checks them against prior code-review outputs, and generates a fragility map to ensure only high-confidence results are highlighted in the final report.

You need dual-agent static analysis for pull requests involving medium-to-large code changes, especially within auth, payments, migrations, or schema files. It is essential when standard automated checks are insufficient to catch complex runtime triggers and missed failure modes prior to merging.

The limitation of using lite mode for finding logic gaps is that it simulates both forward and backward perspectives without spawning agents. While faster, this simulation may lack the deep analytical confidence of a full dual-agent run, potentially missing complex runtime triggers in highly sensitive code.