How we review

We test our own skills and everyone else's the same way, always against doing the task without the skill.

Last updated 2026-10-03

The test

  • We pick a fixed set of real tasks the listing claims to help with, and say how many in the test line (for example "12 paused tasks").
  • We run each task twice in the same app and version: once with the listing, once without any skill. That second run is the baseline.
  • Each run is scored Worked, Partly worked or Did not work. We publish the counts, not a blended score.
  • The test line always states the app, app version, listing version and date.

What we judge, separately

  • Quality: did it do the job well?
  • Task fit: is it the right tool for the job it claims?
  • Safety: what can it read and change, and does it ask before changing anything?

Compatibility labels

Only our tests set "Tested" or "Adaptation needed" for an app. Community reports add detail but never change the label. Everything else is "Not tested".

Our own work

Skills made by our team are marked Dugout-created and go through exactly the same test.

Submissions

We read submitted files. We never run uploaded packages. Duplicates are pointed to the existing listing.