Foreman
Software-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
Overview
Foreman places Jev as a fast supervisor above slower coding agents. Given a ticket, specification, bug report or other software job, it runs Codex, OpenCode, Hermes or LangChain's Deep Agents Code as workers, or attaches supervision to Pi sessions. The coding agent does the engineering; Foreman independently assesses the work against the original job.
From factory evidence (worker diffs, logs and tests) Jev answers Nouls grouped into responsibilities: completion (implementation complete, requirements satisfied, ready to finish), verification (tests sufficient, needs verification) and worker health (stuck, off track, meaningful progress). A Python policy applies these to decide whether to continue, verify or bring in a human, and additional checks can be configured. The README illustrates the output as probabilities per question.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | implementation_complete / requirements_satisfied / ready_to_finish | yes · no |
| Yes / no | tests_sufficient / needs_verification | yes · no |
| Yes / no | worker_stuck / work_off_track / meaningful_progress | yes · no |
How it works
- Give Foreman a job and start a worker
- Collect diffs, logs and test results
- Jev answers responsibility questions
- Python policy continues, verifies or escalates to a human
Strengths
- Independent check on agent claims
- Works with several worker agents
- Configurable responsibilities
Caveats
- Requires a TypeSafe key
- Policy thresholds need tuning per workflow
Getting started
See the README for installing Foreman and running a job with a Codex or OpenCode worker.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- CLI supervisor
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Software factory: sits above Codex workers and has Jev independently judge whether an implementation is complete, its tests sufficient, or a human is needed.
yibie/awesome-jevSoftware-factory loop: Codex implements; Jev independently judges completeness, tests, and whether a human is needed.
AnotiaWang/awesome-decision-modelsAn independent supervisor loop that reads worker diffs, logs and tests, asks Jev Nouls about stuck / off-track / verify, then applies a Python policy.
logicrw/awesome-jev-projectsSoftware-factory supervisor that uses Jev to keep coding agents on task.
cobanov/awesome-jevSoftware Factory Foreman: an agent supervisor that uses Jev decisions to keep coding agents on task.
hellogumbo/awesome-jevPuts Jev as a fast supervisor above slower coding agents such as Codex, starting from a ticket, spec, or bug report.
MrJev/awesome-jev
Frequently asked questions
What does Jev judge?
Completion, requirements, test sufficiency, need for verification and worker health.
Which workers are supported?
Codex, OpenCode, Hermes and Deep Agents Code, plus Pi supervision.
Who decides the next step?
A Python policy using Jev's probabilities.
Related projects
GitHub Action where an AI agent turns an issue into a PR behind guards, checks, a reviewer and a Jev gate that escalates when unsure.
ProjectprototypeTypeScriptDesktop app running coding agents in parallel git worktrees, with optional Jev judgments of each changed file against the task.
AppbetaClaude Code plugin with rules and checks that make agents prove their work; an optional Jev judge triages review findings.
PluginbetaMarkdown + shellClaude Code skills that wait for AI pull-request reviewers and judge whether their reports actually permit a merge, never inferring a pass.
ProjectbetaAsynchronous System-1 supervisory plugin for Hermes Agent with typed Jev decisions, ranking, verification and an opt-in tool gate.
ProjectbetaPythonExperimental Foreman derivative where Jev supervises a Codex worker via Vercel AI Gateway, with programmer-chosen acceptance commands.
ProjectprototypePython