Small decisions. Interesting possibilities.Submit contentSubmit

How it uses Jev

The Pi extension makes one Score question per enabled skill. The task description is shared state, while each skill's name and description are its instructions. It ranks the returned values, applies a configured floor, and loads selected skill files.

What Jev decides

Skill relevance requestExample answers · not a recorded Jev response · Source ↗
Question 1 · skill 0
YOUR APP
INSTRUCTION

How useful is this one skill to an autonomous coding agent doing the task?

STATE

The task description is shared state. Each question adds one skill name and shortened description; skill_0 denotes the first skill in a batch starting at offset zero.

JEV · SCORE
02

App workflow

  1. Prepare the prompt

    Strip the generated skills catalog from the effective system prompt while keeping installed skills available.

  2. Rank applicable skills

    Send the task as shared state and ask a separate Score question for each enabled skill.

  3. Load selected instructions

    Apply the score floor and configured limit, then return full instructions for loaded skills and list other skills above the floor.

Why it is interesting

It removes Pi's generated skills catalog from repeated model requests and replaces it with an on-demand ranking tool; Jev evaluates each skill independently without a keyword prefilter.