Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

dsh-jev-interceptor

DeepSeek Harness plugin where Jev classifies risk, irreversibility, task fit and injection for every tool call before it runs.

Pluginbeta
View on GitHub ↗

Overview

dsh-jev-interceptor wires Jev into DeepSeek Harness decision points where an LLM is overkill and rules are blind. Before a non-read-only tool call runs, one request classifies its risk class (a Choice) and asks Nouls about irreversibility, task match and prompt-injection suspicion. Confident high-risk calls are denied, ambiguous ones escalate to a human, and clearly granted, reversible ones are auto-approved behind argument-evidence gating, saving approval clicks.

When an @session snapshot is injected, a per-message Score rates each message's value to the citing task, so an error traceback survives the byte budget instead of being dropped by age. Every doubt, timeout or missing key degrades to stock behaviour, and nothing can widen a permission. The README contrasts this with DSH's experimental generative auto-review and cites about $0.00002 per request. According to the list summaries, it includes shadow mode with a /jev-stats command and 64 tests.

What the decision model decides

TypeQuestionAnswers
ChoiceWhat risk class does this tool call fall into?—
Yes / noIs this call irreversible / does it match the task / is it suspected injection?yes · no
ScoreHow valuable is this recalled message to the citing task?—

How it works

  1. Intercept each non-read-only tool call
  2. One Jev request classifies risk and asks the Nouls
  3. Deny confident high-risk, ask on ambiguous, auto-approve clearly granted reversible calls
  4. Fail closed to stock behaviour on any error

Strengths

  • Per-call judgment at very low cost
  • Smarter session recall than oldest-first dropping
  • Cannot widen permissions
  • Shadow mode and stats

Caveats

  • Specific to DeepSeek Harness
  • Tool call details are sent to TypeSafe

Getting started

Install the dsh-jev-interceptor npm package as a DSH plugin and set a TypeSafe key.

At a glance

Type
Plugin
Status
beta
Language
TypeScript
Runtime
DeepSeek Harness plugin
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

  • Coding agents: DeepSeek Harness plugin where a Jev `Choice` risk class plus `Noul` irreversibility, task-match, and injection checks gate every non-read-only tool call (deny confident high-risk, ask ambiguous, delegate the rest), `Noul` scope and reversibility questions auto-approve clearly-granted calls behind argument-evidence gating, and a per-message `Score` re-ranks what a referenced session keeps instead of oldest-first dropping — fail-closed to stock behavior, shadow mode with a `/jev-stats` command, 64 tests.

    yibie/awesome-jev

  • ⚡ Millisecond System-1 judgement for every tool call in DeepSeek Harness…

    wh000wh000/awesome-jev-live

  • ⚡ Millisecond System-1 judgement for every tool call in DeepSeek Harness — Jev-powered risk classification & evidence-gated auto-approval. Fail-closed by construction. dsh 生态第一个 System-1 决策插件.

    hellogumbo/awesome-jev

  • Judges every tool call and recalled message inside DeepSeek Harness before it runs.

    MrJev/awesome-jev

Frequently asked questions

What does it judge?

Each tool call's risk, irreversibility, task fit and injection suspicion, plus the value of recalled messages.

Can it auto-approve?

Yes, clearly granted reversible calls with supporting argument evidence.

What if Jev fails?

It falls back to stock DSH behaviour.

Related projects

  • Agent Chaperone

    MCP proxy and hooks that screen agent tool calls before they run and their results before the agent reads them, using Jev judgments.

    MCP serverbetaTypeScript
  • agent-guardrail-jev

    Xano backend template where Jev judges each agent tool call as safe, needs review or dangerous, and uncertain calls wait for a human.

    ProjectbetaXanoScript
  • construct-auto-classifier

    Effect-based safety gate for coding-agent shell commands: fast rules first, then Jev decides allow/deny plus eleven risk checks.

    ProjectbetaTypeScript
  • jev-axi

    Agent-ergonomic Jev CLI and PreToolUse hook that scores shell commands for hazards and screens fetched text for injection.

    PluginbetaTypeScript
  • jev-firewall

    PreToolUse firewall for Claude Code and Codex: shell-aware rules first, then Jev returns allow, ask or block, failing closed.

    HookbetaTypeScript
  • jev-guard (leepokai)

    Security hook giving Codex, Copilot, Gemini, Cursor, pi, OpenCode and ACP an auto-mode-like guard via three Jev questions.

    LibrarybetaTypeScript