Agent tooling
Give coding agents a shared Jev decision layer for review evidence, change risk and next steps, while the host keeps control of files, tools and final actions.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
This minhgv project is distinct from other jev-mcp repositories. The server receives bounded context instead of reading the repository itself. Its v0.3 contract carries evidence and limitations, and the CI helper records non-blocking shadow results. An auto label does not grant permission to merge or deploy.
Ask a developer to start with the project’s deterministic mock mode: fixed test answers without a service key. Use examples stripped of private information. Your coding tool gathers proposed changes, requirements and test evidence. This server judges that bundle rather than reading files itself. Show when evidence is missing or shortened, even if the answer sounds confident.
Collect suggested results beside your existing checks without letting them stop work. Automatic acceptance and blocking remain disabled until tested and explicitly approved. Code checks, tests and access controls still matter. A model’s auto label is advice, not permission to merge changes, run commands or publish them.