Review the system around the model.
Mesh Sentinel reviews dense AI system packets — context architecture, behavior specs, workflows, diagnostics, and eventually code — to detect contradictions, missing operating rules, and behavior drift risks before humans approve the system.
Sentinel detected a cross-document contradiction: one artifact requires explicit resident confirmation before submission, while another instructs automatic submission when fields are complete.
Sentinel modes
The same coherence layer can be entered from different project starting points. Mode 3 is the current working demo.
Spark-to-System
Start with a product spark, create an Intelligence Hub, then monitor the build for drift.
Spec-Driven Development
Add a lightweight Intelligence Hub above build specs and check implementation coherence.
Context Architecture Review
Review multi-document context architecture packets for contradictions and approval blockers.
Repo Coherence Review
Compare existing repos against intended behavior, specs, and architecture decisions.
Demo flow
The current proof of concept runs as a repeatable local demo command and produces structured, human-readable, and traceable artifacts.
What Sentinel found
The sample packet intentionally includes a cross-document contradiction to demonstrate architecture-level coherence review.
Contradiction
The system overview says the assistant must not submit without explicit resident confirmation. The workflow rules say the assistant should submit automatically when all required fields are complete.
Approval blockers
Sentinel converts the contradiction, missing confirmation definition, urgency criteria gap, and auto-submission drift risk into approval blockers.