Content Intake Router

Content Intake Router

Core Principle

Route content by asking: does this need to enter the user's first-order judgment system?

Use the agent for information compression, retrieval, comparison, extraction, and action conversion. Keep the user's own attention for value judgment, taste, strategic intuition, emotional resonance, and identity-shaping learning.

Do not start by summarizing unless the user explicitly asks for a direct summary. First decide the route.

Intake Workflow

1. Fetch source material before routing.

• For any external URL, social post, article, video, podcast, web page, Feishu document/minutes link, or forwarded link, use fetch-skill first to obtain the original text, transcript, clean Markdown, or structured content before making a routing judgment.
• Do not route from a title, preview card, URL slug, or user hunch when the original can be fetched.
• If the user provides a local attachment, screenshot, pasted transcript, or local file, read that source directly; use fetch-skill only when extra external retrieval is needed.
• If fetch fails or access is blocked, state what is missing, route from the available fragments, and mark the confidence as limited.
• For long content, inspect the structure and representative sections before routing.

1. Identify the content's true function.

• News/update
• Tool/tutorial
• Method/framework
• Strategic judgment
• Taste/aesthetic/expression sample
• Emotional/values/person-shaping material
• Current-project evidence

1. Choose one of three routes:

• AGENT代读: Agent reads and converts into summary, decision, action, or archive.
• 人机共读: Agent pre-reads, selects high-value original fragments, user reads only those fragments, then user reaction is turned into judgment/action.
• 本人亲读: User should personally read/watch/listen because the original experience matters.

1. Produce the routing result before any deep processing.

Fetch Contract

When content acquisition is required, the default sequence is:

fetch-skill → source inspection → route decision → optional summary / actions

Use the fetched original as the evidence base for 亲读指数, routing reasons, and recommended next output. For video or audio, prefer original transcript over platform description. For X/Twitter, long articles, YouTube, WeChat articles, and ordinary web pages, rely on fetch-skill's URL-type detection and fallback chain.

Only skip fetch-skill when the complete source has already been provided inside the conversation or as a local attachment.

Routing Rubric

Use AGENT代读 when most signals are true:

• The content is mainly information, news, procedures, prices, feature changes, or tool usage.
• The user's goal is to know "what it says", "whether it matters", "what to do", or "which option is better".
• The content is replaceable; missing the original style or rhythm has little cost.
• The likely output is a checklist, comparison, digest, note, task, or project artifact.

Use 人机共读 when most signals are true:

• The content may affect the user's current projects or evolving judgment, but full reading is not obviously necessary.
• It contains a method, framework, argument, or example that may connect with the user's work.
• The user has a hunch that "this seems important" but cannot yet say why.
• The best workflow is: agent compresses, user reads selected original fragments, then agent crystallizes the user's reaction.

Use 本人亲读 when any strong signal is true:

• The content may change the user's long-term judgment framework, strategy, values, or priorities.
• The user wants to learn the author's expression, taste, structure, tone, or way of thinking.
• The original wording, narrative rhythm, emotional force, or intellectual discomfort is part of the value.
• The topic is tightly coupled to a core project, such as AI coach, personal operating system, organization control tower, business strategy, or content IP.
• The user says they feel strongly hit, excited, uneasy, resistant, or "说不清但觉得重要".

Personal Reading Score

Always assign 亲读指数 from 0 to 5:

• 0: No need to read. Agent can ignore or archive only.
• 1: Agent summary is enough.
• 2: Agent should read and extract actions; user only sees conclusion.
• 3: Human-agent co-reading. User should read selected fragments.
• 4: User should personally read the key sections.
• 5: User should personally read the whole piece or watch/listen to the full original.

Required Output Format

Use concise Chinese output:

路由结论：AGENT代读 / 人机共读 / 本人亲读
亲读指数：0-5

1. 这份内容的本质
...

2. 为什么这样路由
...

3. 如果您不亲自读，会错过什么
...

4. 推荐处理方式
...

5. 下一步输出
...

For AGENT代读, include:

• 3-7 条核心结论
• 与用户当前项目/问题的关系
• 可执行动作
• 是否值得进入长期记忆、项目日志、待办、选题库，或仅临时参考

For 人机共读, include:

• 3-5 个最值得亲自看的原文片段或时间点
• 每个片段为什么值得用户亲自看
• 用户读完后要回答的 1-3 个问题
• 暂缓最终结论，等用户反馈后再沉淀判断

For 本人亲读, include:

• 亲读范围：全文 / 重点章节 / 具体时间段
• 阅读时要带着的问题
• 不建议先看 agent 摘要的原因
• 读完后让 agent 帮用户做的沉淀动作

Output Routing

After route selection, decide what the content should become:

• 不处理: 噪音或低价值内容。
• 临时摘要: 只用于当下理解。
• 行动清单: 变成任务、检查项、采购、验证步骤。
• 项目输入: 进入具体项目的 PRD、项目日志、handoff、研究材料。
• 长期判断: 进入用户的判断框架或原则。
• 内容素材: 进入文章、朋友圈、小红书、演讲或课程素材库。
• 亲读任务: 提醒用户亲自读指定片段，并等待主观反应。

Do not claim that something has been written to memory, logs, docs, or task systems unless the user explicitly asked for that write action and it has actually been completed.

Anti-Patterns

• Do not treat every forwarded content item as worth reading.
• Do not summarize first and route second.
• Do not overuse 本人亲读; user attention is scarce.
• Do not overuse AGENT代读; some content trains judgment only through direct exposure.
• Do not confuse "important information" with "important experience".
• Do not replace the user's reaction with the agent's interpretation.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: quzhi-ai
• Source: quzhi-ai/content-intake-router
• License: MIT

Install and usage instructions live in the source repository linked above.