Jev Agent Toolkit

A portable Agent Skill that teaches coding agents to build with Jev as a bounded judgment layer, plus an optional MCP bridge. One canonical SKILL.md works in Claude Code, OpenAI Codex, Cursor and other Agent Skills-compatible hosts, with no per-host forks.

Install the skill directly from this GitHub repository with your Agent Skills-compatible host. The MCP bridge is separate: it is published to npm as jev-agent-toolkit-mcp, so npx -y jev-agent-toolkit-mcp is the regular setup path, verified on 2026-09-20 including the handshake.

Jev makes bounded semantic judgments with a probability attached; the host agent's LLM reasons, writes text and orchestrates; deterministic code handles arithmetic, counting, dates, parsing, builds, thresholds and control flow; external tools act in the world. It deliberately does not route everything through Jev.

No. The skill and the MCP bridge are separate, and the README says most users only need the skill. MCP is described as an interoperability layer for reaching Jev through a tool call - a transport, not Jev itself.

Capability-driven workflows for code, browser research, Blender and Unreal Engine, built around Jev as a bounded judgment layer. The README's diagram shows the portable skill core feeding a lead agent or orchestrator that drives code, browser and other tools.

CI and protocol tests run against a local mock upstream, and there is no automated live TypeSafe test. Manual synthetic live checks passed on direct HTTP, the Python SDK, the JavaScript SDK and the local MCP bridge, with end-to-end MCP runs in Claude Code, Codex and Cursor - one-off manual runs, no load or stability testing.

reiswaffel78 publishes it, listed under logicrw, with CI running on its workflows and the MCP package published to npm as jev-agent-toolkit-mcp. The digest metadata lists no license field, so check the repository before relying on reuse terms.