jev-kit
Claude Code kit wiring Jev into a tool-call guard, sub-agent sizing, file search, a browser agent, review, belay and compaction.
Overview
jev-kit bundles what is needed to run Jev with Claude Code: a tool-call guard, a tier guard that right-sizes sub-agents, file search, a browser agent, review, a belay component, compaction, monitoring and installers. Its premise is that the judge is the easy part; the work is deciding which tool calls to question, which sub-agent rung a task needs, what to redact before sending, and how to fail open when no answer arrives.
The README reports that with Jev choosing each click, the browser agent's Claude bill was $0.0008 per run versus $0.1868 with Sonnet choosing, at the same 9 of 9 success, and links to how the numbers were taken. The installer never runs sudo, never writes outside the home directory and never edits settings.json unless --wire is passed; otherwise it prints the JSON to add. All components read one TYPESAFE_API_KEY from a kit-level config file, and a Python installer covers native Windows.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Should this tool call be questioned or blocked?(inferred) | — |
| Choice | Which sub-agent tier does this task need?(inferred) | — |
| Choice | Which element should the browser agent click next?(inferred) | — |
How it works
- Run install.sh (or the Windows installer)
- Add TYPESAFE_API_KEY to the kit config
- Wire hooks into Claude Code with --wire or manually
- Jev judges tool calls, sub-agent tiers, searches and clicks; fail open on errors
Strengths
- Large reported cost reduction for the browser agent
- Conservative installer
- Many integrations in one kit
Caveats
- Browser result is a 9-run measurement
- Many components to understand
Getting started
`git clone https://github.com/jonathanavis96/jev-kit.git ~/code/jev-kit && cd ~/code/jev-kit && install/install.sh`.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- Claude Code hooks, daemon and tools
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Everything you need to run TypeSafe's Jev with Claude Code: a tool-call guard, tier guard, file search, browser agent, review, belay, compaction and installers.
logicrw/awesome-jev-projectsEverything you need to run TypeSafe.
wh000wh000/awesome-jev-liveEverything you need to run TypeSafe's Jev with Claude Code: a tool-call guard, tier guard, file search, browser agent, review, belay, compaction and installers.
hellogumbo/awesome-jevClaude Code bundle: a tool-call guard, sub-agent sizing, and Jev wired into search, browsing and review.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide in the kit?
Which tool calls to question, which sub-agent tier fits, file search relevance and browser clicks.
Will the installer change my settings?
Only with --wire; otherwise it prints what to add.
What savings were measured?
Browser-agent Claude cost fell from $0.1868 to $0.0008 per run at the same success rate.
Related projects
Experimental browser-only MCP server where Jev picks a code-owned action ID and a human approves before execution.
MCP serverprototypeTypeScriptPre-action gate and browser agent where rules decide first, then Jev or a local Jev-compatible model, emitting allow, confirm or block.
ServiceresearchPythonVercel AI SDK layer that uses Jev to auto-approve, ask a human or block each tool call, and to prune irrelevant tools before a turn.
LibrarybetaTypeScriptClaude Code job-application tool that finds roles, fills forms in Chrome, reads every answer back and sends only those you approve.
ProjectbetaTypeScriptOne-command npx installer for a set of Jev-powered agent tools: context compaction, browser agent, prompt enhancer and skill router.
ProjectbetaJavaScriptLibrary, CLI and MCP server where an LLM states each step's outcome and Jev decides element, action, value and done/blocked/irreversible.
MCP serverbetaTypeScript