JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Architecture and workflow guide

How to test and rank saved instructions for AI assistants

You will learn how to safely test a system that helps an AI assistant choose which saved instructions to use next. The guide explains how to check choices silently before letting the software act on them.

Original by DicklesworthstoneEvaluation

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits

“skillranker” by Dicklesworthstone. Read the original source.

This expanded guide is an AI-narrated adaptation prepared by JevMade. It expands the source’s essential ideas, examples and caveats in JevMade’s own words and is not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

Our summary

SkillRanker is a program that helps an AI assistant find the right saved instructions for its current task. When an assistant has too many instructions to choose from, this software narrows down the list so the assistant does not get confused or waste time reading useless information.

The system looks at the recent conversation and the available instructions. It sends this information to Jev, an AI tool that chooses from options rather than writing an answer. Jev compares the choices, including an option to pick nothing, and suggests the best match for the next step.

This setup requires a private access code for the Jev service and an active internet connection. It does not run its own AI models locally. The guide is useful for people who want to test how well their assistant selects instructions before letting it make those choices automatically.

Key takeaways

  1. Review what information the software will send over the internet before you allow it to connect.
  2. Record how the system chooses instructions silently before you let it give advice to the assistant.
  3. Test the selection process using saved examples to see how it handles failures and missing options.

The creator of the software reported these behaviors and measurements. Reviewers looked at the code and instructions but did not run the program to test the claims.

Repository README and documentation · Source reviewed

Read the original guide Opens the author’s site in a new tab.