Video unavailable. Open the source below.
NewJev Review
An MCP tool supplies repeated software-quality measurements to coding agents.
Added to Jevfast
How it uses Jev
The MCP tool accepts the task, focused diff, selected files, and repository context explicitly supplied by the coding agent; it does not read the repository itself. Jev receives those fields as state and answers per-metric typed Noul applicability, Score, and Choice weakness questions. The server converts the answers into metric scores, confidence, prioritized rubric hints, and local comparisons; the coding agent diagnoses causes, changes code, checks it, and rescores.
What Jev decides
Illustrative Jev decisionExample answers · not a recorded Jev response · Source ↗
Question 1 · correctness applicable
YOUR APP
INSTRUCTION
Is there enough relevant evidence in the supplied software change to assess correctness?
STATE
A coding task and focused diff, with any files and repository context supplied by the coding agent.
JEV · NOUL
YesNo
Question 2 · correctness score
YOUR APP
INSTRUCTION
Rate correctness and requirement fit for the supplied software change.
STATE
A coding task and focused diff, with any files and repository context supplied by the coding agent.
JEV · SCORE
0
Question 3 · correctness weakness
YOUR APP
INSTRUCTION
Which correctness weakness matters most in the supplied change?
STATE
A coding task and focused diff, with any files and repository context supplied by the coding agent.
JEV · CHOICE
- no_material_issue
- missing_behavior
- edge_case
- incorrect_assumption
- regression_risk
App workflow
- Supply focused context
The agent sends a task and diff, adding relevant file contents or repository context when needed.
- Build metric questions
For each quality dimension, create Noul applicability, Score, and weakness Choice questions with defined criteria.
- Evaluate and transform
Validate Jev’s typed answers, convert the zero-based score to a displayed 1–10 value, and create priorities and rubric hints for applicable metrics.
- Improve and compare
The agent inspects weak dimensions, makes a justified change, validates it, then resubmits the current context with the previous evaluation for local delta comparison.
Why it is interesting
It uses typed scalar feedback as a repeated review loop while reserving diagnosis and code changes for the coding agent. Noul applicability and a predefined weakness Choice qualify which score feedback is actionable.