360 Secure SkillHub
360 安全技能中心

brand knowledge base

核心功能是将零散品牌材料整合为标准化、可复用的品牌唯一真实来源(SSOT),支撑下游内容、销售、客服、RAG等工作流。支持上传原始资料或填写intake表单两种输入方式,首次使用需告知必填资料。输出包含多格式品牌知识库、FAQ、分析报告等,资料不全时会标记“待确认”项并列出跟进问题,适用于企业统一品牌信息的场景。

ljseeking
ljseeking
热度 27.63
最近更新 2026-05-20 04:13:34
分类:知识依赖:代码执行 / 读写文件 / 内部调用大模型应用场景:数据处理 / 摘要总结 / 模板 / 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 brand-knowledge-base
Metadata
name
brand-knowledge-base-builder
description
Build a reusable brand knowledge package from uploaded brand documents and/or a filled intake form. Use when the user wants a standardized brand SSOT containing company intro, products/services, target users, selling points, cases, FAQ, contact info, website links, competitors, banned words, and brand tone.

Brand Knowledge Base Builder

Purpose

Use this skill to turn scattered brand materials into a standardized, reusable brand package that can feed downstream content, sales, support, FAQ, RAG, or Agent workflows.

This skill is for building a Single Source of Truth, not for writing one-off marketing copy.

What It Accepts

Use one or both of these input modes:

  1. Upload raw materials

    • 官网文案
    • 公司介绍
    • 产品资料
    • 销售话术
    • FAQ 草稿
    • 客户案例
    • 访谈或语音转文字
  2. Fill an intake form

    • Run the skill in template mode to generate intake_form.md, intake_form.yaml, or intake_form.json
    • Let the user fill key fields first, then append raw materials if needed

First-Use Requirement

On the first turn this skill is used, do not jump straight into extraction.

You must first clearly tell the user:

  1. This skill builds a reusable brand SSOT, so the company needs to provide source materials first.
  2. The company can provide materials in either of these ways:
    • Upload raw materials directly
    • Fill the intake form first, then optionally append raw materials
  3. If the company does not provide enough information, the skill may still generate a draft, but missing fields must be marked as 待确认, and the analysis report must list follow-up questions.

You must explicitly ask the user to provide the following minimum required materials:

  • 公司基础信息:品牌名称、公司名称、所属行业、产品类别、一句话定义、100 到 300 字公司简介
  • 官网与公开链接:官网首页、产品页、案例页、FAQ 页;如果没有完整官网,至少提供对外介绍页或文档
  • 产品与服务说明:卖什么、解决什么问题、怎么交付、如何收费
  • 目标客户信息:核心客户、次级客户、决策人、使用者、不适合客户
  • 核心卖点:至少 3 条卖点,并尽量附证据、数据、客户反馈或资质依据
  • 联系方式:销售邮箱、支持邮箱、电话、微信/社媒、地址、工作时间、预约方式
  • 品牌语气与表达规范:语气关键词、必须出现的说法、禁用词、禁用承诺、安全替代表达
  • 合规边界:哪些能说、哪些不能说、哪些效果不能承诺、哪些内容必须人工审核

You should then strongly recommend these additional materials:

  • 客户案例 2 到 5 条:客户类型、原问题、解决方案、结果、证据
  • FAQ 种子问题 10 到 20 条:价格、效果、安全、交付周期、合作方式
  • 竞品与替代方案:直接竞品、间接竞品、传统做法、外包方案、行业 SaaS
  • 销售或客服话术:开场白、异议处理、报价说明、边界说明
  • 原始素材:官网文案、宣传册、访谈纪要、录音转文字、产品说明、案例文档

Required Coverage

The final package should cover:

  • 公司简介
  • 产品与服务
  • 目标客户
  • 核心卖点
  • 使用场景与客户案例
  • FAQ
  • 联系方式
  • 官网链接
  • 竞品与替代方案
  • 品牌语气
  • 禁用词与禁用承诺
  • 合规边界

Workflow

  1. Intake

    • On first use, explicitly tell the user what the company must provide
    • Collect files and/or a filled intake template
    • Prefer mixed mode when the user has partial structure plus raw docs
    • If the user has not prepared enough material, direct them to the intake template instead of proceeding silently
  2. Core extraction

    • Produce a canonical JSON package
    • Prioritize factual sections over derivative writing assets
  3. Analysis

    • Score completeness
    • Identify missing fields, conflicts, assumptions, and follow-up questions
  4. Asset generation

    • Generate FAQ, glossary, standard messaging, and llms.txt inputs
  5. Export

    • Render the full package into:
      • brand_knowledge_base.json
      • brand_knowledge_base.yaml
      • brand_knowledge_base.md
      • faq.md
      • glossary.md
      • standard_messaging.md
      • llms.txt
      • analysis_report.md

Safety Rules

  • Do not invent missing facts.
  • Use conservative wording.
  • Strip or rewrite unverified absolute claims.
  • For regulated industries, keep or add strong disclaimers in compliance_boundary.
  • Human review is still required before publishing or loading into production systems.
  • If the user has not yet provided the minimum required materials, be explicit about what is missing before building the package.

Files To Read

  • For intake guidance: prompts/intake.md
  • For extraction rules: prompts/extraction.md
  • For analysis rules: prompts/normalization.md
  • For derivative assets: prompts/asset_generation.md
  • For schema shape: templates/brand_knowledge_base.json

Runtime Notes

  • main.py is the executable entrypoint.
  • --generate_intake_template creates fillable intake files.
  • --input and --inputs accept raw material files.
  • --intake_file accepts a structured intake form.
  • --render_from_json re-renders the full artifact set from an existing package JSON.

First Reply Pattern

When the skill is first invoked, the assistant should lead with a message equivalent to:

使用本 Skill 前,请先按要求提供品牌资料。至少需要:公司简介、官网链接、产品服务、目标客户、核心卖点、联系方式、品牌语气与禁用词、合规边界。强烈建议同时提供客户案例、FAQ、竞品信息、销售话术和其他原始素材。你可以直接上传资料,或者先填写 intake 表单。若资料不完整,系统会生成初稿,并把缺失项标记为“待确认”。