WEBVTT

00:00:00.460 --> 00:00:03.313
An agent directory is a starting point for research.

00:00:03.433 --> 00:00:06.204
A feature listed on a page is not a test result.

00:00:07.360 --> 00:00:11.135
Start with the job, the tool surface, and where the agent runs.

00:00:11.255 --> 00:00:13.924
Those requirements make the comparison meaningful.

00:00:15.093 --> 00:00:19.108
Keep documented claims separate from capabilities reproduced in a test.

00:00:19.228 --> 00:00:21.268
Record the source and date for each claim.

00:00:22.427 --> 00:00:25.825
A missing field means the catalog does not establish the answer.

00:00:25.945 --> 00:00:28.206
It does not prove the feature is unsupported.

00:00:29.360 --> 00:00:32.755
Local execution and offline operation are different claims.

00:00:32.874 --> 00:00:37.621
Check model, service, and network dependencies before treating them as equivalent.

00:00:38.793 --> 00:00:41.986
Test shortlisted tools on the same representative task.

00:00:42.106 --> 00:00:46.432
Record the version, environment, output, and any limits you observed.

00:00:47.593 --> 00:00:50.988
Use AgentList to narrow the field and follow its source links.

00:00:51.108 --> 00:00:54.119
Verify important claims against the work you need done.