Bringing Ops Inspections to Research Groups: Using WorkBuddy for Automated Paper Resource Checks

Last week, I saw a post about WorkBuddy operations practice, where the author used AI to assist in setting up server inspections, triggering alerts when CPU, memory, or disk exceeded limits. I agree with a sentence in there:

This isn't an article about AI automating everything automatically. The entire process involved me stating requirements, AI proposing solutions, and me making decisions and verifying.

I'm a master's student in my third year and have been using WorkBuddy for about a month. I previously wrote about its boundaries, and this time I tightened them a bit more. Initially wanting to slack off, I ended up forced to straighten out naming and permissions.

The lab's shared drive contains four types of things: paper drafts, raw data, literature PDFs, and plagiarism check reports. Previously, I checked manually every week, mainly fearing three things: papers changed but plagiarism reports weren't rerun, data tables missing columns, and shared drive permissions getting messed up. These aren't big issues, but hitting any one means late-night revisions. So instead of doing server monitoring, I treated paper materials like a server needing inspection and set up a weekly automated health check.

Setting up a weekly automated health check with WorkBuddy is straightforward. Open the WorkBuddy web version; on my left side, there's Automation. If not, find Tasks and click New Automation. Automation executes instructions at fixed times. I selected Sunday 20:30 weekly. I chose the weekend because data is mostly settled, and my advisor might look at it on Monday. In the task description, paste a prompt: read the research group's shared drive path, check paper materials, generate an inspection table listing filename, last modified time, anomaly type, and handling suggestion. Select Tencent Docs Spreadsheet for output, create a new document named "Paper Materials Inspection Weekly Report." For notifications, only let it generate a Lark draft. Think of Lark as a collaboration platform like Feishu; drafts are safer than sending directly to group chats. After saving, a new task appears in the list with status "Waiting to Execute." Opening the spreadsheet the next day shows the results.

I wrote four rules like this. Backup expired: If the main paper file hasn't had a new backup in over 7 days, mark as anomaly. Plagiarism check expired: If the plagiarism report's modification time is earlier than the paper's latest modification time, mark as anomaly. Data anomaly: If CSV files have missing columns, excessive null values, or inconsistent group fields, mark as anomaly. (CSV is a spreadsheet file.) Literature drift: If the difference between Zotero export count and paper reference count exceeds 3, mark as anomaly.

After the first run, it indeed listed the files, but there were two types of misjudgments. It didn't recognize versions from local renames; plagiarism report filenames were too messy, so old versions were identified as new ones. Later, I standardized naming: Main paper files use PaperName_Version_Date, and WorkBuddy's recognition became much smoother.

Permissions were also set separately. WorkBuddy has read-only inspection rights, no write-back, no deletion. Only my advisor and I can edit the core paper directory; peers have read-only access. Peers can comment on the inspection weekly report sheet but cannot modify original files. Read-only boundaries are more important than prompts.

After running for three weeks, it didn't write the paper for me, but turned missed checks into recorded anomalies. Week 1 found the main file newer than the plagiarism report; Week 2 found a data table missing a group column; Week 3 found two extra accounts in permissions, resulting in an extra anomaly line in the report.

For daily maintenance, I only do three things: check paths monthly, keep one inspection sheet per week (store max four weeks). Don't dump full sensitive papers into public tools; only provide filenames, dates, and table fields.

Don't let WorkBuddy auto-modify files right away. Initially, I wanted it to rename files casually and almost mixed my advisor's annotated version with the clean version. Later, we established that AI only checks, leaving core decision-making to humans.

If you want to set this up, don't start complex. First, let it check one table: Which files haven't moved in the last 7 days? Which files moved but weren't synced? Once that works, add plagiarism, data, and permissions checks. Define boundaries clearly, and WorkBuddy becomes a reliable inspector.