home / skills / hexbee / hello-skills / research-content-router

research-content-router skill

/skills/research-content-router

This skill orchestrates research to content production, delivering publishable conclusions with evidence chains and reusable material libraries.

npx playbooks add skill hexbee/hello-skills --skill research-content-router

Review the files below or copy the command above to add this skill to your agents.

Files (3)
SKILL.md
5.0 KB
---
name: research-content-router
description: 将“研究 + 内容生产”任务按 A/B/C 分级调度,先收敛受众/目标/载体/角度,再走 Research→Synthesis→Content 流水线,输出可发布内容与证据链。适用于需要结论可信、结构稳定、可复用素材沉淀的写作任务。
---

# Research Content Router

## 何时使用

当用户提出以下需求时触发本技能:

- 研究某个主题并产出内容(推文、长文、简报、脚本等)
- 需要“结论 + 证据链”而不是纯资料堆
- 需要在“传播、可信度、转化”之间选择内容方向
- 希望把研究产出沉淀为可复用素材

## 系统目标

- 用最少确认次数拿到可发布内容
- 将研究转化为“结论 + 证据链 + 可复用素材库”
- 先收敛选题/角度/受众/格式,再写作

## 1) 任务分级(A/B/C)

### A 级:先确认再执行

满足任一即 A:

- 需要联网检索、引用来源、或对外发布研究结论
- 涉及观点立场、攻击性评价、法律合规、投资建议
- 要求长内容(>800 字或 >10 条推)或系列内容
- 需要用户私域数据(笔记/文档/项目/客户信息)

默认动作:

1. 先给 3 个方向 + 推荐 1 个
2. 用户选定方向后再执行

### B 级:自主执行后返回

适用:

- 主题、受众、格式已明确
- 任务以扩写、改写、结构化、提炼为主

默认动作:

1. 直接执行
2. 返回时附带:
   - 1 行“采用的写作假设”
   - 3 个可替代角度(便于快速改向)

### C 级:静默归档

适用:

- 链接清单、摘录、术语解释、资料暂存
- 用户未要求“立即发布”

默认动作:

1. 打标签
2. 写摘要
3. 归档到素材库(不主动打扰)

## 2) Router 六变量(必须补齐)

研究/内容任务统一补齐以下变量:

1. 受众:写给谁
2. 目标:相信/行动/讨论/收藏
3. 载体:推文线程/长文/简报/脚本/演讲稿
4. 角度:反常识/方法论/案例拆解/对比评测
5. 证据级别:一手源必须/可选/不需要
6. 尺度:长度、数量、交付时间

执行规则:

- A 级:将 1-6 转成可选项让用户点选
- B 级:缺失项可自行假设,但需明确写出假设
- C 级:只存档,不追问

## 3) 三段式流水线

### Stage 1: Research(搜集)

必须输出:

- 结论候选 3-5 条
- 每条结论对应证据(链接/原话/数据)
- 证据分级:
  - `S` 一手源(财报/论文/官方文档/原始数据)
  - `A` 权威二手源(主流媒体/机构报告)
  - `B` 一般二手源(博客/社区)
  - `⚠️` 未证实来源(传言/未核实自媒体)

硬规则:

- 没有证据的观点不得进入 Stage 2

### Stage 2: Synthesis(收敛)

必须输出:

- 核心论点(1 句话)
- 支撑点 3 条(每条都要挂证据)
- 反方风险 1 条(强制项)

硬规则:

- 缺少“反方风险”则判定失败,必须重做

### Stage 3: Content(生产)

默认交付:

- 主版本(按选定载体)
- 备选标题 5 个
- 结尾 CTA 3 个(提问/收藏引导/讨论引导)
- 可复用素材:
  - 金句 10 条
  - 类比 3 个
  - 图表建议 2 个

## 4) A 级确认卡(固定三选一)

当用户说“研究某主题/写内容”时,先返回:

- 方向 A(更快传播):反常识结论 + 强观点
- 方向 B(更强信任):证据链 + 一手源引用
- 方向 C(更能转化):方法论/清单/框架,可直接照做

并附一句推荐:

- 默认推荐 B:研究情报任务中,可信度决定复用价值

## 5) 原子能力(9 个)

1. `Query Builder`:主题扩展为检索式(中英/同义词/时间窗)
2. `Source Triage`:按 `S/A/B/⚠️` 去噪分级
3. `Evidence Extractor`:抽取关键引用与数据(带出处)
4. `Claim Ledger`:结论-证据对照表
5. `Contrarian Risk`:反方观点与失败条件
6. `Outline Generator`:按载体产出结构
7. `Hook Factory`:批量生成标题与开头
8. `Compression`:长内容压缩成推文/短文/口播
9. `Repurpose`:研究一稿多用(推文→长文→脚本→邮件)

要求:

- 每个能力都要固定输入/输出格式,保证可堆叠

## 6) 交付格式(强约束)

### 研究交付(必选)

- TL;DR(<=80 字)
- 结论(3 条)
- 证据(每条结论 2-4 条,含来源等级)
- 风险(>=1 条)
- 可写角度(3 个)

### 推文线程交付(必选)

- 标题 5 个
- Thread(8-12 条,每条 <= 280 字)
- 引用/出处标注(需要时)
- CTA 3 个

## 7) 指令模板(可直接复用)

```text
主题:{X}
受众:{创业者/开发者/投资人}
目标:{收藏/讨论/转发}
载体:{推文线程 10 条}
证据级别:{必须一手源/可选/不需要}
约束:{语气/禁忌/长度}
输出:按“研究交付 + 推文线程交付”
```

## 执行守则

- 永远先分级,再决定是否提问
- 永远先收敛变量,再进入写作
- 永远保留反方风险,防单边结论
- 永远保持“可发布 + 可追溯 + 可复用”

Overview

This skill organizes combined research and content production tasks into an A/B/C triage and runs a Research→Synthesis→Content pipeline to deliver publishable outputs with traceable evidence. It forces early convergence on audience, goal, format, angle, evidence level, and scale so writing proceeds from a stable brief. The result is conclusions plus evidence chains and a reusable asset library for future reuse.

How this skill works

The router first classifies the request as A (confirm then execute), B (execute then report assumptions), or C (archive silently) based on risk, need for sources, length, and data access. It fills six mandatory variables (audience, goal, carrier, angle, evidence level, scale), runs three stages—Research (collect conclusions with graded sources), Synthesis (one-line core argument, three evidence-backed supports, one contrarian risk), and Content (final deliverable plus titles, CTAs, and reusable micro-assets). Hard rules prevent progression without documented evidence and a stated counter-risk.

When to use it

  • You need publishable content grounded in evidence rather than raw notes.
  • The task must balance credibility, reach, and conversion (e.g., promotion vs. trust).
  • You want research outputs condensed into reusable assets (quotes, analogies, chart ideas).
  • Deliverables require traceable citations or one‑hand source verification.
  • You need a predictable workflow for multi-format repurposing (thread → longform → script).

Best practices

  • Always supply or confirm the six router variables before deep work to avoid rewrites.
  • Treat A-level tasks as interactive: present three directions and recommend one.
  • Require at least one contrarian risk in synthesis to surface failure modes.
  • Label every evidence item with S/A/B/⚠️ and never advance claims without sources.
  • Produce a small reusable asset pack (golden lines, analogies, chart suggestions) with each deliverable.

Example use cases

  • Research a market trend and publish a data-backed long article with source links and reusable quotes.
  • Turn interview notes into an evidence-backed 10-tweet thread with CTAs and alternate angles.
  • Archive a list of references and excerpts as C-level material for later content mining.
  • Create a short playbook (methodology) that converts research findings into actionable steps for customers.

FAQ

What triggers A/B/C classification?

Risk factors like public publishing, legal/financial stance, need for primary sources, content length, or use of private data shift tasks to A; clear briefs with editing needs are B; simple archives or lists are C.

What minimum evidence is required to proceed?

Every candidate conclusion must have 2–4 supporting items labeled S/A/B/⚠️; claims without evidence are blocked from Stage 2.