Skip to content
Projects09 / AI verification / developer tooling

Yashkhou Verify

Yashkhou Verify is a model-agnostic verification suite for checking AI work with deterministic evidence rather than trusting the system that produced it.

StackJavaScript · CSS · HTMLStatusPublic · early releaseYear2026LicenseMIT
Concept artwork for Yashkhou Verify
VERIFY / conceptual system artwork
01THE PROBLEM

Why this needed to exist.

AI can generate files, code and actions quickly, but a claimed success is not the same as a verified outcome. The proof layer needs to stay independent from the model that created the work.

02THE APPROACH

How I built around it.

  1. 01Split verification into three independent tools for artifacts, repositories and actions.
  2. 02Prefer deterministic assertions, executable tests and source evidence before probabilistic judgement.
  3. 03Emit machine-readable evidence and non-zero exits so verification can gate larger workflows.
  4. 04Treat irreversible actions differently from reversible ones and preserve rollback or compensation paths where possible.
03WHAT SHIPPED

Capabilities

  • Artifact checks for XLSX, PPTX, DOCX and PDF
  • Repository acceptance gates
  • SHA-256-addressed evidence
  • Approval and idempotency primitives
  • Rollback and compensation flows
  • Hash-chained local action journal
  • CLI and CI-friendly failure signals
  • Independent npm packages
04SYSTEM

Architecture

Runtime
Node.js 20+
Artifact layer
Deterministic file assertions
Repo layer
Executable acceptance gates
Action layer
Approval · rollback · compensation
Evidence
SHA-256 + local reports
License
MIT
05 / PRINCIPLES

What the system is meant to preserve.

  • The verifier should not need to trust the generator
  • Deterministic checks before model judgement
  • Evidence should survive outside the chat that created it
  • Recovery paths are part of safe execution
06FAQ

Useful answers.

Is Yashkhou Verify one package?

No. The repository contains three independent tools: Artifact Verify, Repo Verify and Action Guard. They can be used separately or combined.

Does it guarantee AI output is correct?

No. It is verification infrastructure, not a guarantee of correctness. The current release focuses on explicit structural, content, code and action checks that can be measured.

Can it be used in CI?

Yes. The repository tools are designed to emit evidence and failure exit codes that can gate automated workflows.

NEXT BUILD / 10

COVERLIST

A commercial launch site for a productized booking-site service aimed at independent hospitality operators.

Keep exploring
Concept artwork for COVERLIST