Tasks

All 40 tasks of task set 2.1: the exact prompt, the inputs and the checklist.

All inputs are synthetic. Test pages live on a separate test site, agent-meter-lab.pages.dev/test/. Files that go into the chat as attachments are linked from each task. The checklists are scored by a person or a script after the run. The answer keys are not published while the task set is in use, so they cannot end up in a run.

IDTaskCategoryInput
T01Quick question with a sourceSmalltest pages
T02Inbox triage, 20 emailsSmallfiles
T03Routine: watch a page (daily)Routinedaily test page
T04Research: 5 facts with sourcesResearchtest pages
T055 PDF invoices to a tableLargefiles
T06CSV cleanup (500 rows)Datafiles
T07Mini site from a briefCodefiles
T08Web form with an errorBrowsertest pages, files
T09Impossible taskResearch / failuretest pages
T10Follow-up in an old threadLargeT06 inputs
T11Connectors loaded, not usedSmalltest pages
T12Good morningSmallprompt only
T13Resume a chat after an hourLargefiles
T14Notes to a to-do listSmallfiles
T15Small change in an existing appCodefiles
T16Trip plan with constraintsLargetest pages
T175 subagents in parallelResearchtest pages
T18Project plan from a briefLargefiles
T19Budget spreadsheet with formulasDatafiles
T20Fix a bug in a small repoCodefiles
T21Write unit testsCodefiles
T22Script: CSV to JSONCodefiles
T23SQL query on SQLiteCode / datafiles
T24Refactor plus READMECodefiles
T25Price comparison, 3 shopsBrowsertest pages
T263-step formBrowsertest pages
T27Extract a table from a pageBrowser / datatest pages
T28Log in and downloadBrowsertest pages
T29Routine: inbox digest (daily)Routinedaily test page
T30Routine: poll every 5 minutesRoutinetest pages
T31Only create a routineRoutineprompt only
T32Routine: daily X digestRoutineprompt only
T33Handoff and resume in a new chatLargefiles
T34Browser with full-page screenshotsBrowsertest pages
T35Long context (30,000 words)Researchfiles
T36Summarize 3 URLsWritingtest pages
T37Long report, 1,500 wordsWritingtest pages, files
T38Generate an imageWriting / mediaprompt only
T39SEO blog post, 800 wordsWritingfiles
T40Analysis plus chartDatafiles

How the runs are measured is on the method page.