A coherence and behavior-drift layer for AI-native projects.

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.

Why: teams lose coherence as AI systems move from idea → specs → context → code.

Sentinel review goals are presets over the same core workflow: source of truth, evidence, review question, report.

Prepare files for handoff

Check architecture, specs, context files, design principles, and operating rules before sending them to devs or stakeholders.

Check Intelligence Hub alignment

Compare tools, repos, and outputs against project principles, source-of-truth strategy, and operating intent.

Check whether a repo matches specs

Review README, code, tests, configs, and implementation snippets against intended behavior and acceptance criteria.

Check drift from approved baseline

Compare new artifacts or commits against an approved packet, spec, prior report, or project baseline.

The hosted demo includes a stable walkthrough, live K2 reviews, document packet review, and repo coherence review.

1. Ingest packet

Read files, pasted artifacts, or selected repo files.

2. Review coherence

K2 reviews contradictions, missing rules, and drift risks.

3. Normalize

The harness validates JSON and normalizes status.

4. Report

Generate human-readable approval blockers and next tests.

5. Trace

Preserve inputs, files, model, outputs, and finding provenance.

Sample test files

The current demo uses a public service request assistant packet with system overview, behavior spec, context architecture, workflow rules, and testing diagnostics. We can expose downloadable sample files here once the copy is final.