Pi Jev by Theo Oliveira
A Pi coding-agent extension uses Jev to discover relevant tools and skills, make typed judgments, and optionally guard tool calls and compaction.
Added to Jevfast
How it uses Jev
The Pi extension uses Jev for semantic tool routing, skill discovery, typed Choice/Noul/Score evaluations, and a typed subagent. Optional guards, compacting, model selection, and post-run gate checks provide additional decision points; automatic behavior is opt-in.
What Jev decides
Illustrative Jev requestExample answers · not a recorded Jev response · Source ↗
Question 1 · is breaking
YOUR APP
INSTRUCTION
Does this change introduce any breaking API changes?
STATE
A code diff supplied as state.diff.
JEV · NOUL
YesNo
App workflow
- Discover candidates
Read the prompt and Pi’s registered tools or skills, then use local keyword matching to build a bounded candidate set.
- Evaluate a typed request
Send state and named Noul, Choice, or Score questions to Jev for tool/skill routing or a direct evaluation.
- Apply optional agent integrations
When enabled, guard tool calls, preserve selected context during compaction, choose a model, or evaluate a post-run result.
- Use or report the result
Activate recommended tools, suggest skills, return typed answers, or report a gate outcome; if Jev is unavailable, avoid blindly activating unjudged tools.
Why it is interesting
It exposes the same fast typed-decision client across agent workflows, with explicit failure handling and opt-in switches. Tool and skill discovery can shortlist candidates locally before Jev evaluates which ones fit the task.