Try a resource on a real task

Use a small, non-sensitive task from your own work. A completed trial helps us learn whether a resource is useful beyond finding it in search.

1. Find your profession and task, then choose your language.
2. Open the instructions and read the required inputs. Replace placeholders with fictional or anonymised facts.
3. Use the resource in your normal assistant. For GitHub skills, use the complete pinned package and its setup guide.
4. Check whether every requested output is present. Verify calculations, claims, source references and preserved names, amounts and dates.
5. Record whether you finished the task, how useful the result was, and how much editing it needed using the review form on the resource page.
6. If you are fluent in the selected language, also check natural wording, regional terms and unchanged facts. Leave that rating blank if you cannot judge.

Taking part in the five-person pilot?

Your moderator will give you a student, professional or business task. Start at the homepage, find a resource, and use it in your normal assistant. Say when you feel stuck. Keep personal and confidential details out of the task. Your moderator records anonymous timings and ratings after you agree.

For independent reviewers

Review outputs without knowing which version produced them. Score completeness, factual accuracy, usability, handling of missing inputs and language quality from 1 to 5. Record specific errors and corrections. Compare a plain task request with the resource-assisted answer, using the same model and inputs. Repeat each condition three times.

Language editions need a fluent reviewer; regulated tasks need an appropriately qualified reviewer. Automated checks are screening tools. A passing string check does not establish a good complete answer.

The pack includes 113 profession tasks, 25 language choices, input requirements, acceptance criteria and instruction revision hashes. It is a test plan, not completed user research.