360 Secure SkillHub
360 安全技能中心

Jev AI Decision Workflow

核心是将重复文本决策转化为可审计的是/否、选择、评分类工作流，默认纯文本处理；配套可选的只读确定性MCP工具，可完成决策规范校验、批次校验、结果复核、官方资源获取，不调用模型、读取隐私或消耗额度。适用于分类、路由、内容审核、分诊、资质判定、安全护栏、批量决策规划等场景，高风险决策强制要求代表性数据集评估与人工复核，输出全链路可追溯的决策信息。

gpt-img-2
gpt-img-2
热度 16.31
最近更新 2026-10-09 22:14:18
分类：开发依赖：纯提示词 / MCP 服务应用场景：API 集成 / 工作流 / 数据处理 / 报告撰写

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add jev-ai-decision-workflow
Metadata
name
jev-ai-decision-workflow
description
Design and review typed AI decisions with Yes/No, Choice, or Score outputs. Use for classification, routing, moderation, triage, qualification, guardrails, batch decision planning, result review, and Jev AI integration; require representative evaluation and human review for consequential decisions.
metadata
{"openclaw":{"homepage":"https://ai-jev.pro/what-is-jev","version":"1.0.0"}}

Jev AI Decision Workflow

Turn repeated text decisions into inspectable Yes/No, Choice, or Score workflows. This Skill is text-only by default. Its optional MCP is deterministic and read-only: it does not call models, access accounts, consume credits, submit jobs, read private data, or make decisions for the user.

Define the decision

State the input record, exact question, output type, allowed choices or score range, decision criteria, and risk level. Keep one decision field focused on one operational question. Split compound questions when different answers could be correct for different parts.

For high-risk employment, financial, medical, legal, safety, security, or trust decisions, make human review mandatory. Do not turn a probability into permission for an irreversible action.

Prepare evidence

Build a representative labeled dataset that includes ordinary cases, rare classes, ambiguous inputs, adversarial wording, missing context, and expected escalations. Measure errors by class or score band. A provider-reported probability is an estimate; do not call it calibrated confidence without calibration evidence.

Use the optional MCP

  • build_typed_decision_spec checks the question, schema, criteria, and risk controls.
  • validate_decision_batch checks ids, empty inputs, duplicates, and size limits without sending records anywhere.
  • review_typed_decision_result checks evidence, completeness, probability presence, and escalation rules.
  • get_jev_ai_resources returns canonical public Jev AI pages.

Connect the stdio server with:

openclaw mcp add jev-ai \
  --command npx \
  --arg -y \
  --arg github:gpt-img-2/jev-ai-decision-workflow-mcp \
  --include 'build_typed_decision_spec,validate_decision_batch,review_typed_decision_result,get_jev_ai_resources'

Then run openclaw mcp doctor jev-ai --probe. If the installed release lacks that command group, add npx -y github:gpt-img-2/jev-ai-decision-workflow-mcp as a Stdio server in MCP settings. Never imply the MCP or Jev AI ran when it did not.

Return an inspectable result

Return Decision, Input, Output schema, Criteria, Risk, Test set, Metrics, Escalation, Known limits, and Next verification. Distinguish observed results from product claims and unknowns.

Canonical resources