What it is
Foreman is a Python asyncio runtime that sits above Codex or OpenCode workers. It takes a free-form software job and independently assesses whether the implementation is complete, tests are sufficient, verification is needed, or a human is required.
How it uses Jev
Jev assesses ten dimensions: implementation_complete, tests_sufficient, requirements_satisfied, needs_verification, ready_to_finish, meaningful_progress, worker_stuck, work_off_track, agents_md_drift, needs_human. Each is a Noul question sent in one request. Probabilities are validated, normalized, stored in state.json, and used by a deterministic Python policy to decide actions like CONTINUE, START_VERIFIER, STEER_WORKER, STOP_WORKER, FINISH, or ESCALATE.
Primitives:noul
Technique worth stealing
Use a fast decision model to independently supervise slow generative workers without interrupting their loop.
Try it
FOREMAN_WORKER_BACKEND=opencode foreman run --repo ./my-project --job "Add request retries"
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricyes0.88
- Shows a System One patternunclear0.44
- Handles uncertaintyno0.33
- Measuredno0.04
- Runnableyes0.62
- Worth recommendingno0.39
- Model replicano0.07
- Problem scopescore on a 0–2 scale1.91
- About Jevyes0.97
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.