research-content-router

Research Content Router

When to Use

Trigger this skill when the user raises the following needs:

• Research a topic and produce content (tweets, long-form articles, briefs, scripts, etc.)
• Need "conclusions + evidence chain" rather than a pure pile of material
• Need to choose a content direction among "reach, credibility, conversion"
• Want to precipitate research outputs into reusable material

System Goals

• Get publishable content with the fewest confirmations
• Transform research into "conclusions + evidence chain + reusable material library"
• First converge topic/angle/audience/format, then write

1) Task Grading (A/B/C)

A-tier: Confirm before executing

Meets any one = A:

• Requires online retrieval, citing sources, or publishing research conclusions externally
• Involves viewpoints/stances, aggressive evaluations, legal compliance, investment advice
• Requires long content (>800 characters or >10 tweets) or a content series
• Requires the user's private-domain data (notes/documents/projects/customer information)

Default action:

1. First give 3 directions + recommend 1
2. Execute after the user selects a direction

B-tier: Execute autonomously and return

Applicable:

• Topic, audience, and format are already clear
• The task is mainly expansion, rewriting, structuring, or distillation

Default action:

1. Execute directly
2. When returning, include:
   • 1 line of "writing assumptions adopted"
   • 3 alternative angles (for quick redirection)

C-tier: Silent archiving

Applicable:

• Link lists, excerpts, terminology explanations, temporary material storage
• The user has not requested "publish immediately"

Default action:

1. Tag
2. Write a summary
3. Archive to the material library (without proactively disturbing)

2) Router Six Variables (must be completed)

Research/content tasks uniformly complete the following variables:

1. Audience: who it is written for
2. Goal: believe/act/discuss/save
3. Medium: tweet thread/long-form article/brief/script/speech draft
4. Angle: counterintuitive/methodology/case breakdown/comparative review
5. Evidence level: primary sources required/optional/not needed
6. Scale: length, quantity, delivery time

Execution rules:

• A-tier: Turn 1-6 into selectable options for the user to click
• B-tier: Missing items may be assumed independently, but the assumptions must be explicitly written out
• C-tier: Only archive, do not follow up

3) Three-Stage Pipeline

Stage 1: Research (collection)

Must output:

• 3-5 candidate conclusions
• Evidence corresponding to each conclusion (link/original quote/data)
• Evidence grading:
  • S Primary source (financial reports/papers/official documentation/raw data)
  • A Authoritative secondary source (mainstream media/institutional reports)
  • B General secondary source (blogs/communities)
  • ⚠️ Unverified sources (rumors/unverified self-media)

Hard rules:

• Viewpoints without evidence must not enter Stage 2

Stage 2: Synthesis (Convergence)

Must output:

• Core argument (1 sentence)
• 3 supporting points (each must be tied to evidence)
• 1 counterargument risk (mandatory item)

Hard rules:

• If "counterargument risk" is missing, it is judged a failure and must be redone

Stage 3: Content (Production)

Default deliverables:

• Main version (according to the selected medium)
• 5 alternative titles
• 3 closing CTAs (question/save prompt/discussion prompt)
• Reusable materials:
  • 10 quotable lines
  • 3 analogies
  • 2 chart suggestions

4) A-level confirmation card (fixed choose-one-of-three)

When the user says "research a certain topic/write content," first return:

• Direction A (faster spread): counterintuitive conclusion + strong viewpoint
• Direction B (stronger trust): evidence chain + primary source citations
• Direction C (better conversion): methodology/checklist/framework, directly actionable

And attach one recommendation:

• Default recommendation B: in research intelligence tasks, credibility determines reuse value

5) Atomic capabilities (9)

1. Query Builder: expand topic into search queries (Chinese-English/synonyms/time window)
2. Source Triage: by S/A/B/⚠️ denoise and grade
3. Evidence Extractor: extract key citations and data (with sources)
4. Claim Ledger: conclusion-evidence comparison table
5. Contrarian Risk: opposing viewpoints and failure conditions
6. Outline Generator: produce structure according to medium
7. Hook Factory: batch-generate titles and openings
8. Compression: compress long content into tweets/short posts/spoken scripts
9. Repurpose: one research draft used multiple ways (tweet -> long article -> script -> email)

Requirements:

• Each capability must have a fixed input/output format to ensure stackability

6) Delivery format (strong constraints)

Research delivery (required)

• TL;DR (<=80 characters)
• Conclusions (3)
• Evidence (2-4 per conclusion, including source level)
• Risks (>=1)
• Writable angles (3)

Tweet thread delivery (required)

• 5 titles
• Thread (8-12 items, each <= 280 characters)
• Citation/source attribution (when needed)
• 3 CTAs

7) Instruction template (directly reusable)

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”

Execution guidelines

• Always classify first, then decide whether to ask questions
• Always converge variables first, then enter writing
• Always retain counterargument risk to prevent one-sided conclusions
• Always maintain "publishable + traceable + reusable"