VFS-Bench 1.0 Tasks
Open a Task to Compare VFS vs MCP.
A/B testing
Read twenty scenarios of clickstream out of BigQuery, fill record.csv with each version's conversion rate, and then act on the winner: create a Cloud Storage bucket prefixed promo-assets-for-b if B wins, or write one Cloud Logging entry if A does.
Academic PDF report
For each paper under study for an AI conference, find the first author's full name, their complete affiliation exactly as printed in the paper (every institution, down to the department where given), and their Google Scholar profile link, and fill them into the Excel sheet in the workspace.
Academic warning
Join latest_quiz_scores.csv against seven historical score tables in BigQuery, write every student whose score fell more than 25% into bad_student.csv, and write a CRITICAL Cloud Logging entry for each one who fell more than 45%.
Add BibTeX
Add five papers, from the Qwen2.5-Math technical report and Lemur to Code Llama and MMLU, to ref.bib in the same format as the existing entries, copying the BibTeX from the OpenReview conference page whenever a paper has a conference version.
Apply PhD email
Find an email from kaiming in the inbox, gather the materials it asks for from the workspace, and reply with them attached under an exact subject line.
Arrange workspace
Sort 22 loose files, from a CV and spreadsheets to photos and music, into a prescribed three-level folder tree.
Canvas exam schedule
It is 16 January 2025: read the course announcements on Canvas and fill exam_schedule.xlsx with every final exam the student still has to sit, leaving out courses that have ended, that they need not attend or that have no final. Strip the "-x" suffix from course codes and names, take the instructor as proctor by default, and write TBD wherever details are not yet released.
Canvas course setup
Find the courses registrar admin3 is responsible for in the master schedule attached to the "Course Schedule Notification" email, and create and publish each one on Canvas. Enroll its instructor as teacher, post an announcement with the instructor's name and class time, publish its syllabus from the workspace, and set up resource sharing where one instructor teaches the same course more than once.
Canvas art history quiz
Create a quiz called "Classic Art History Questions" in the Art History course on Canvas, add the four given one-point multiple-choice questions on Caravaggio, Courbet, Duchamp and Rothko versus Newman, and mark the correct answer to each.
Unfinished Canvas quizzes
Complete the student’s unfinished Canvas quizzes.
Canvas Python homework grading
Download the homework2 Python files attached to the submission emails, run each one, and grade it on Canvas: 10 if it runs correctly, 0 otherwise, counting only each student's latest submission. The requirements are in assignments/homework2.md and the students' Canvas IDs in student_canvas_ids.csv.
Canvas new student notification
Find the students who changed majors and are listed on the new Introduction to AI-8 summary sheet in the workspace but not yet enrolled, enroll them in the course on Canvas, and privately message only them that their first assignment grade will match their second and that they should complete it diligently.
Canvas late work submission
Submit the student’s existing assignment files and email the TA.
Cooking guidance
Read a list of ingredients on hand, pick three dishes out of a 322-recipe Chinese home-cooking corpus that use as many of them as possible, write the dish names into cuisine.json, and write everything still missing into a shopping.csv.
Course assistant
Find the NLP students who have not submitted their final presentation, checking the course spreadsheet against the inbox, and email each of them with their name and student ID.
Course schedule
Pick this student's classes for the semester out of a timetable and a master exam table, then write the exams they will sit to exam_schedule.jsonl, sorted by when they start. Who the student is — their id, their department, their English score — is in Memory.
Courses TA homework
Find every file belonging to Assignment 3 of "Operating Systems Fundamentals", rename them all to Name-College-StudentID-OS-HW3 using the college/student-ID correspondence table, sort them into os_hw3/C, os_hw3/Rust and os_hw3/Python, and delete the original code files — including those from other courses and other assignments.
CVPR postdoc search
Using the student's profile in personal_info.md, find the three researchers who published the most papers at CVPR 2025 and meet their requirements for a postdoctoral advisor, and write their names one per line to top3_match_researchers.txt.
Dataset license issue
Someone filed an issue asking which licence two released datasets carry. Find the sources each was built from, take the most permissive, reply in a fixed format, close the issue, and append the same line to both dataset cards on the Hub.
Detect revised terms
Read new_law.pdf and a set of case judgment documents, and identify every cited clause whose substantive content differs from the corresponding provision in the new law — changes to the rule, scope, condition or legal effect. Clauses merely renumbered or relocated do not count, and only clauses the judgment actually quotes are in scope.