The idea

John Yeo describes adding Jev ahead of a Slack agent. The classification step uses the request to identify a suitable skill, tool and parameter choices before handing work to the agent, reducing the amount of discovery inside the agent loop.

Inside the decision loop

01 / CONTEXT

Slack request + available skills

02 / DECISION

Select the skill and tool

03 / ACTION

Hand the selection to the agent

What to take from it

Give the router a short, current catalog of capabilities. Validate selected parameters against the tool schema and leave room for the agent to reject a poor match.

KEEP IN MIND

The author reports an improvement in their setup. That does not establish the same improvement for other skill catalogs, workloads or permissions. Tool authorization still belongs in application code.

Follow the original work

This case is an editorial interpretation of a public community demonstration. It has not been reproduced or benchmarked by this publication.

Original post by @johnyeo_