Foreman
An experimental supervisor uses Jev to assess and steer a coding agent.
Added to Jevfast
How it uses Jev
Foreman serializes factory observations and asks a batch of Noul checks keyed by active responsibilities. Python parses the judgments and policy decides whether to continue, steer, stop, retry, verify, or finish while the worker runs.
What Jev decides
Illustrative Jev decisionExample answers · not a recorded Jev response · Source ↗
Question 1 · implementation complete
YOUR APP
INSTRUCTION
Does the current implementation satisfy the completion check?
STATE
Serialized factory observation, including current worker and verification evidence.
JEV · NOUL
YesNo
App workflow
- Collect factory events
Observe worker lifecycle, output, tests, and repository instructions.
- Build checks
Select active responsibility checks and turn each into a Noul question.
- Assess with Jev
Send the observation and all active checks in one request.
- Apply policy
Parse judgments and choose an intervention or allow work to continue.
Why it is interesting
A fast classifier supervises coding work asynchronously without taking over the worker's individual tool choices.