JevFlow
Claude Code plugin where Claude plans phased work with checks and, at each stop, Jev judges whether the work is really done.
Overview
Jevflow is a Claude Code plugin aimed at four failure modes of coding agents: declaring victory early, drifting off the goal or losing it after compaction, multiple agents tripping over each other, and stopping on crashes or rate limits. For any real task Claude lays out a small plan of phases, each with a definition of done and a check. Every time Claude tries to stop, Jevflow runs the phase checks and asks Jev where the work stands: which phase is it in, is the phase finished, is Claude stuck or off the goal, or claiming done too early.
A small, fully tested policy then decides to block with a concrete next step, advance to the next phase, or allow the stop and archive the flow. The rule is that Jev informs and code decides: a probability alone never advances a phase or completes a goal, a failing check always blocks, and if Jev is unreachable it falls back to checks only. Every stop is journaled so crashed or compacted sessions resume where they were without repeating side effects, a supervisor restarts hung sessions, several agents can share one plan, and a live web viewer shows all flows. MIT licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is the current phase finished? | — |
| Yes / no | Is Claude stuck or off the goal? | — |
| Yes / no | Is Claude claiming done too early? | — |
| Choice | Which phase is the work currently in? | — |
How it works
- Claude writes a phased plan with checks
- On each stop, run phase checks
- Ask Jev about phase, completion, stuck or off-goal
- Policy blocks, advances or allows stop; journal for resume
Strengths
- Checks always outrank the model
- Resumes after crashes or compaction
- Multi-agent shared plans with live viewer
Caveats
- Requires Claude Code
- Quality depends on the checks Claude writes
At a glance
- Type
- Plugin
- Status
- beta
- Language
- TypeScript
- Runtime
- Claude Code plugin
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A Claude Code plugin that keeps your AI agents honest. Claude plans tasks as phases with checks, and when it tries to stop, Jevflow runs them and asks Jev if the work is really done. Many agents, one plan, with a live viewer to watch it all.
logicrw/awesome-jev-projectsA Claude Code plugin that keeps your AI agents honest. Claude plans tasks as phases with checks, and when it tries to stop, Jevflow runs them and asks Jev if the work is really done. Many agents, one plan, with a live viewer to watch it all.
hellogumbo/awesome-jev
Frequently asked questions
Can Jev alone mark a task complete?
No, a probability never completes a goal on its own; checks always outrank the model.
What if Jev is unavailable?
Jevflow falls back to checks only.
Can multiple agents work together?
Yes, several agents can share one plan, visible in the live viewer.
Related projects
Claude Code Stop hook that blocks an unverified done: one four-question Jev call only when files changed and no check passed.
PluginbetaTypeScriptStop hook for coding agents where Jev checks each attempted stop against plain-language rules and sends the agent back if one is violated.
HookbetaPythonClaude Code Stop hook that uses Jev to catch turns ending half-done: no action, broken promises, unwatched work or wrong language.
HookbetaPythonGitHub Action where an AI agent turns an issue into a PR behind guards, checks, a reviewer and a Jev gate that escalates when unsure.
ProjectprototypeTypeScriptSoftware-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
ProjectbetaPythonAsynchronous System-1 supervisory plugin for Hermes Agent with typed Jev decisions, ranking, verification and an opt-in tool gate.
ProjectbetaPython