xienda/dsh-jev-verify
Jev (TypeSafe AI System One) decision model as DeepSeek Harness agent tools, with an opt-in auto-guard and local usage panel.
dsh-jev-verify brings Jev, TypeSafe AI's System One decision model, into DeepSeek Harness as agent tools: given a state plus typed questions it returns typed answers with calibrated probabilities in one parallel API call. It adds jev_decision (batch up to 25 typed questions), jev_choose (rank 2-10 options by fit and risk), jev_verify (run a frozen 27-question labeled benchmark against the live API), plus read-only guard and usage panels. An opt-in auto-guard performs risk and loop checks before shell-like tool calls, and the plugin reports measured local usage and cost rather than provider credit.
Install
dsh plugin --profile web add dsh-jev-verifyCopied verbatim from the README's install line (npm package dsh-jev-verify). The README states it requires Node >= 20 and dsh >= 0.1.5-rc.2.
Compatibility
The README states it requires Node >= 20 and dsh >= 0.1.5-rc.2; engines.dsh is advisory. It exposes agent tools and an opt-in auto-guard.
Details
- Repo: xienda/dsh-jev-verify
- Category: Agent Capabilities
- Stars: 1
- Version: GitHub repo xienda/dsh-jev-verify; npm package dsh-jev-verify at 0.8.2 (registry check 2026-10-06). The README notes the version deployed in its own testing was 0.7.3 and requires dsh >= 0.1.5-rc.2.
FAQ
- Does Jev generate explanations?
- No. The README says Jev scores, it never explains - it returns typed answers with calibrated confidence, and the reasoning stays with you.
- What does Jev need to run?
- The README states Node >= 20 and dsh >= 0.1.5-rc.2, and the plugin calls the live Jev API (an optional local quota can hard-stop before the call).
Alternatives
tt-a1i/archify · PM-Shawn/Abu-Cowork · openma-ai/open-managed-agents