06 · The Library: MD / CSV / HTML

What it fixes

AI output has a chronic problem: it scatters and is never seen again.

This task's report is in task-20260304, last week's is in task-20260228, and a colleague's copy is on their laptop. To reuse anything you first have to find it, and then hand it to the agent all over again.

The Library (资料库) is where humans and agents both keep their output. It closes the loop: agent output files itself, team material lives in one tree, tasks can pull Library content back in for further work, and a finished page can become a link other people can open.

Open it from the sidebar.

Not a cloud drive

It is not a pile of files, it is a set of references

The Library lets MD, CSV and HTML reference each other: documents cited by pages, tables rendered by pages, change one thing and it changes everywhere.

That is the essential difference from a cloud drive. A drive stores snapshots; the Library stores live relationships.

Three capabilities

#	Capability	The problem it solves
01	Content management	Personal and team material organised separately; scattered files into one tree; read and edit directly from a task
02	Multi-person, multi-agent collaboration	Team spaces and permissions; MD review with AI revision; collaborative HTML editing
03	Lightweight publishing	Turn local HTML into an online link; build small apps from HTML plus CSV

Five core properties

Humans and agents share permissions. The Library is native to WorkBuddy, and when an agent reads or edits Library content during a task, its permissions are exactly yours. What you cannot see, it cannot see.

One tree holds everything. The AI trio (MD / CSV / HTML), office and local files (PDF / Word / PowerPoint / Excel), images, audio and video, all in one place.

Personal and team are separated. "My documents" for personal output and saved items, "Team space" for shared project material. Everything has a clear home.

Changes are a loop. Add from the Library into a task → ask the agent to work on it → save the result back. Material is not a one-off upload; it enters a continuous working cycle.

AI edits are reviewable and reversible. An agent's changes to an MD document do not overwrite the original — they appear as pending, and only get written in once you accept.

The MD / CSV / HTML split

The most transferable idea in the Library's design: separate data from presentation.

Form	Job	Role	Typical use
MD	Decides content	Narrative	Proposals, weekly reports, explanations — text and structure
CSV	Stores data	The single source of truth	Lists, schedules, inventory, performance — data you maintain over time
HTML	Handles expression	Presentation	Reads the CSV and renders charts and lists, turning a static page into a data-driven one

CSV as the single source of truth is held in a WorkBuddy online table that several people can maintain simultaneously; the HTML page reads it by table ID and renders live.

One line to remember

Change the content in MD, the data in CSV, the look in HTML — none of them interfere.

The value shows up in maintenance: numbers change without rebuilding the page, styling changes without touching the data. On a one-off you feel nothing; three months in it is obvious.

A concrete example

Say you want a team weekly dashboard:

• weekly-notes.md — what the dashboard is for, how the metrics are defined, who updates it
• weekly-metrics.csv — one row per week, the numbers. Anyone can edit, edits take effect immediately
• dashboard.html — reads weekly-metrics.csv, renders a line chart and some cards

Next week's numbers arrive: add a row to the CSV. The page updates itself and the explainer never changes.

Getting started

1. Open the Library from the sidebar to find "My documents" and "Team space".
2. Put agent output in. Either have the agent save it directly during a session, or click share on an artifact and turn on publish-as-website, which publishes and files it in one move.
3. Upload local files, office documents and media directly to the tree, or clip a web link.
4. When you want the agent to work from existing material, pick a space or document in the Library and click Add to task (添加到任务).
5. Save the new output back to the Library for later reuse.

Publishing as a website

Publish as website is off by default

Once on, anyone with the link can open it. You turn it on deliberately and you can turn it off any time.

Published links use the workbuddy.link domain, and the publisher is responsible for the content. Content violating law or platform rules is not permitted.

Before enabling, check two things: is the content suitable for the public, and is the CSV data it reads suitable for the public? The page may be clean while the data source is not.

Turn it off when you no longer need it. Not fastidiousness — a link that has spread cannot be recalled.

Permission boundaries

• Content in "My documents" is visible only to you
• Team space content is visible to space members, scoped by the space's permission tier
• An agent only reads what the current user may access — never other members' or spaces' material
• Publish as website is off by default

Check content before adding it to a task

Before adding a space or document to a task, make sure it contains nothing unsuitable for a model context.

A staff roster with ID numbers is fine for you to read. Once it enters the conversation context it is part of the model's input.

When adding members by invite link, make sure the link only goes to people who should join.

Storage

Content	Stored where
Uploaded office and local files (PDF / Word / PowerPoint / Excel, images, audio, video)	Tencent Drive
MD, CSV and HTML created inside WorkBuddy	WorkBuddy cloud

The Library shows used and total capacity. Uploads check free space first and prompt you to clear files or expand if there is not enough.

Credits

Free: storage, organising, searching, manual editing.

Charged: having WorkBuddy process, analyse or generate content, plus AI revision, revision-from-comments, and HTML select-and-edit.

In collaboration, whoever clicks pays

Credits are billed to whoever actually initiated the AI operation: the person who clicked and sent the instruction.

Take revision-from-comments: the comment text is only reference context for the AI, and the credits go to whoever clicked revise-from-comments and sent it, not to whoever wrote the comment.

Accepting or rejecting a suggested revision costs nothing.

Four recommendations

1. Save important output back to the Library promptly so it does not vanish when a session expires.
2. Divide team spaces by project to avoid mixing things up.
3. Write content in MD, keep data in CSV, present in HTML — lowest long-term maintenance cost.
4. Use inline comments and review before editing shared documents so you do not overwrite someone else's work.

You judge the accuracy

Whether agent output is accurate and appropriate is your call, and reviewing before formal use or external sharing is advised. The Library makes sharing very easy, which is both the benefit and the risk.

FAQ

Is the Library the same as project assets?

Project assets are the Library as it appears in a project — the Assets tab on the project page. Team spaces and project assets share the same underlying capability.

How do I know when an agent edited my document?

An agent's edits to MD documents appear as pending changes and do not overwrite the original. They are written in only when you accept. That is the biggest difference from ordinary "let the AI edit my file".

Does Library content enter every task automatically?

No. A project library is injected as reference material into tasks inside that project; otherwise you add content to a task explicitly.

Next

07 Automations on a schedule — let the recurring work run itself.