Published to npm as llm-checker@3.8.1.
This GitHub release synchronizes the ModelVet 3.8.0 feature release and the 3.8.1 attribution patch.
Highlights:
- Structural verification for GGUF and safetensors through ModelVet WebAssembly, fully local and offline.
- New verify command with machine-readable JSON and CI-friendly verdict exit codes.
- Fail-closed Ollama verification gates for installed --verify and ai-run --verify, with the explicit --allow-unverified override.
- Structural validation policies and audit reporting.
- MCP verify_model support plus multi-client setup for Claude, Codex, Cursor, Windsurf, Gemini, Kimi, Grok, and generic clients.
- Correct published MCP executable and npx launch command.
- Prominent credit to ModelVet, created by Tetsuo AI, with the bundled MIT notice documented in THIRD_PARTY_NOTICES.
Validation:
- 53 of 53 tests passing.
- GitHub CI green on Node 18, 20, and 22 plus Windows UI and policy gates.
- Zero production dependency audit findings.
- Published npm tarball verified to include README and THIRD_PARTY_NOTICES.
An ACCEPT verdict confirms structural validity only; it does not assess model behavior, provenance, or poisoned weights.
Full details: changelog.