COMMUNITY OBSERVEDMetric claims: AUTHOR REPORTEDSOURCE AUDITED🤖 Jev Agent

jev-rules

Jev picks which of your rules apply, so Claude only sees those.

Original Claude Code session demo
Original Claude Code session demo. Author's animated Claude Code session showing rule selection and scores.

Overview & result

A Claude Code plugin asks Jev whether each project rule applies to the current prompt and passes only selected instructions to the agent. The repository includes a real-session animated demo and a fail-open path that preserves all rules on error.

From the original post

Jev picks which of your rules apply to each prompt, so Claude only sees the ones that matter.

— Elia Alberti

How Jev fits in the loop

  1. Keep project rules outside the repeated agent prompt.
  2. Ask Jev a Noul question for each rule against the current request and file activity.
  3. Inject selected rules into Claude Code; on error, deliver every rule rather than silently omit one.

How to reproduce

  1. Install the documented Claude Code plugin and enable its rules pane if supported.
  2. Define several standing rules and submit prompts that should match only a subset.
  3. Inspect the visible rule scores and verify that a failed Jev call falls back to showing all rules.

Why this build matters

Shows a precise decision boundary for long-lived coding instructions: Jev selects relevance, while the host controls what the agent actually receives.

Limitations

  • The plugin uses an early-access Claude Code feature for its optional in-app rules pane.
  • Rule selection can miss a relevant instruction; the documented failure path shows all rules instead of dropping them.

Patterns