Is this still maintained or should I start new work on it?

No. The author discontinued it and directs new work to jevcore, which is maintained, harness-independent and offline by default. This repo stays available under MIT for its design ideas.

What does Jev actually decide inside the harness loop?

At pre-step it restricts tool categories, at pre-execute it grades specific calls as allow, ask, hold or deny, and at request time it picks a verified model route. Retry loops use a deterministic guard.

Can I try the examples without a TypeSafe key or network?

Yes. Defaults are mock provider plus shadow mode, fully offline with synthetic answers and logged decisions. Live TypeSafe calls need an explicit key and still transmit state in shadow mode.

What happens on errors, timeouts or newer harness versions?

Errors, timeouts and stale state fail closed to ask or hold, never allow, and later policies cannot widen earlier restrictions. Newer harness releases are unverified and issues are no longer worked.

Which version should I install, and what is the trap?

0.1.4 is the last release and the latest dist-tag for both @buberlo/dsh-jev and @buberlo/jev-core. Do not install 0.1.2 or 0.1.3: they carry a literal workspace:^ version that fails with EUNSUPPORTEDPROTOCOL, and 0.1.3 was abandoned after a staged-version conflict. Verified against DeepSeek Harness 0.1.6-alpha.2 with @deepseek-ai/cordis 4.0.2 and @typesafe-ai/sdk 0.6.0.

What are the mode and provider switches, and what do they leak?

Two independent switches. Mode controls whether answers may act: off makes no requests, shadow logs decisions without changing behaviour, enforce applies them. Provider controls where answers come from: mock is deterministic synthetic scenarios with no network and is the default, while live hits the TypeSafe API and needs an explicit apiKey. Live plus shadow still transmits state to TypeSafe; it only skips applying the decisions.

What does the test suite actually cover?

147 tests, 85 in the core and 62 DSH integration, plus 25 evaluation fixtures and an on-prem support regression set. The headline measurement is a protected audit trail blocked in 10 of 10 runs, with method and limits written up in docs/benchmark.md and docs/evaluation.md. The comparison demos are clearly labelled replay with no live replanning, and they include one documented false positive.

CommunityJev Agents & MCP

Dsh Jev — buberlo

Jev decision layer for DeepSeek Harness that narrows visible tools, gates risky calls, guards retry loops and routes models, with Jev permitted only to restrict or gate, never to widen access.

See highlights

Open github.com

Stars: 25
Language: TypeScript
License: MIT
Published by: Community
Author: buberlo
Last verified: Oct 2026

buberlo/dsh-jevArchived

User repository on GitHub

View on GitHub

Jev-powered decision layer for DeepSeek Harness

Forks: 8
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

This repository stays available under the MIT licence. If any of these ideas are useful to you, please take them — in jevcore or anywhere else: Monotone permission narrowing. Jev may only restrict, never widen: tool selection intersects with existing restrictions, and call assessments compose as deny > hold > ask > allow after calling next(), so later policies are never skipped. Errors, timeouts and stale snapshots fail closed (ask/hold, never allow). See docs/policy.md.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Monotone permission narrowing: assessments compose deny over hold over ask over allow.
Shadow mode plus offline mock default logs decisions without changing behavior.
Harness-independent core plus plugin, with coding, ops-router and game examples.
Versioned fixtures plus with and without Jev benchmarks with documented methods and limits.

Watch out

Development is discontinued with no further releases and only tested against older DeepSeek Harness alpha builds, so newer hosts may break. For maintained Jev support the author points to jevcore instead. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

Other listings of this project

Forks and copies of this project in the directory. This entry is the canonical one.

dsh-jev

Plugin suite pairing Jev decision primitives with DeepSeek Harness for tool pruning, loop blocking, risk guarding, skill routing, and result shaping.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

zhangxaochen/dsh-jevTypeScript

dsh-jev

DeepSeek Harness bundle that registers a jev_ask tool plus a jev skill so a DSH agent can send mixed Noul, Choice, and Score questions to Jev.

Jev Agents & MCP#community#jevbest#catalog-import

Community

noetion/dsh-jevTypeScript

dsh-jev

DeepSeek Harness community plugin that classifies each turn's newest user message through the systemone endpoint and injects the verdict as tagged runtime context for tone and priority.

Jev Integrations & Adapters#community#github

Community

Excalibur9527/dsh-jevJavaScript

More Jev resources like this

duet-agent

Agent harness for work that outlives one chat, with resumable turns, routed state machines, and observational memory across sessions and processes.

Jev Agents & MCP#community#jevbest#catalog-import

Community

dzhng/duet-agentTypeScript

skilled-agent-harness_spec-driven-loops

Assistant framework that stores specs, memory, and specialist agents as plain markdown files in your repo for Claude Code, Codex, Opencode, and similar runtimes.

Jev Agents & MCP#community#jevbest#catalog-import

Community