WorkBuddy's most efficient use: standardizing daily reports into templates
Community Discussion · Tracks

WorkBuddy's most efficient use: standardizing daily reports into templates

Classmate ZhouClassmate ZhouSep 92026/09/09 58 views

Messy data is manageable; the trouble is explaining the same set of fields every day. Last week I read a tutorial on WorkBuddy going from 0 to 1, and a sentence in it was quite right: don't give the entire large directory containing privacy, finance, and customer contracts directly to the Agent. I've used WorkBuddy for about a month, and now I generate daily order reports with basically just three mouse clicks.

My environment is backend development, and I'm used to local files, so this might not apply to everyone. When using WorkBuddy for the first time, don't start with complex requirements. After logging into the client, the main interface is roughly divided into Conversation Area, Task Area, and Function Area. Switch to Daily Office in the Conversation Area, and entries like Document Processing and Data Analysis & Visualization will appear below. Beginners can start with Data Analysis & Visualization.

First, create a directory; don't pile files in the default directory. I strictly use workbuddy/daily-report/, placing only the day's anonymized orders.xlsx and refund.xlsx inside. Original contracts, customer tables, and financial details do not go in. When selecting the workspace, point to this directory. Afterwards, the Agent reads and processes files only within this scope. If the path is given too broadly, it really will find things it shouldn't.

When creating a new task, don't write "Help me make a daily report" in the input box. That's too vague, and the output looks like a meeting. I strictly write the following: Read orders.xlsx and refund.xlsx in the current workspace, join by order_id, output daily_report.xlsx, keeping fields date, region, amount, refund_amount, net_amount, retaining two decimal places for amounts, and output abnormal rows separately to exception_rows.xlsx. On the first run, I verify column names and results. Once verified, I save this prompt as a template named daily-report-v1.

Afterwards, every day I just put the new files into daily-report/, select the template, and click run. The output lands in the workspace, so I don't have to dig through the conversation area. This action is very much like a commit message: fixed input, fixed output, less explanation.

Regarding permissions, I made one rule: templates are read-only. Colleagues can drop tables into inbox/ but cannot change mappings. Field definitions are clearly written in Feishu Docs, and WorkBuddy only runs daily-report/. If it errors, first check if fields were renamed. For example, if amount is changed to Amount (in Chinese), the join will return empty. My handling is to revert to the original table; I don't let the Agent guess on the fly. If it guesses once, I might have to write two extra lines of validation, which is unnecessary.

For daily operations, I only do a few things: rename output files with dates every day and keep exception tables; check weekly if fields in the template have changed; if someone requests adding a column, update the template description first, then update the WorkBuddy template. Don't temporarily add a line like "also calculate gross margin" in every conversation. Adding a temporary line makes the result look like a meeting, and no one knows how the final outcome was derived.

Office Agents truly save time by running against the same template every day. I later solidified the review checklist with this mindset too. I previously wrote about WorkBuddy reconciliation templates; looking back, the paths are the same: smaller directories, stricter fields, shorter templates. Leave the rest to it to run. As long as it works, that's enough.

1 replies

?
Ctrl + Enter to reply
Engineer Jiang

This saves effort on writing daily reports, but does it mask the real power wall issues? When Unisoc was doing basebands, templating just meant missing key debugging details.