Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

github-star-classifier-jev

Script that classifies your starred GitHub repos into GitHub Lists with Jev, dry-run by default and with a confidence threshold.

Projectprototype
View on GitHub ↗

Overview

GitHub Star Classifier sorts your starred repositories into GitHub Lists using Jev and gh api graphql. It is safe by default: runs are dry-run unless --apply is given, you can test on a small sample or a single repository, and missing managed lists can be created as private. Re-classification replaces only memberships in its managed taxonomy lists and preserves manual lists. According to the list summary, it also helps review stars worth dropping.

A confidence threshold (0.50 by default) leaves results below it untouched and reports them. A compare option asks a frontier LLM the same two questions through TypeSafe's system-one-adapter and prints its answer under Jev's; Jev still decides, and a summary shows how often the two agree. It requires Python 3.13+ and an authenticated GitHub CLI.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich list in the taxonomy does this starred repository belong to?(inferred)—
Yes / noIs this star worth dropping?(inferred)yes · no

How it works

  1. Fetch starred repositories via gh
  2. Ask Jev the classification questions
  3. Skip results below the confidence threshold
  4. Apply list memberships with --apply

Strengths

  • Dry-run by default
  • Preserves manual lists
  • Optional comparison with an LLM

Caveats

  • Requires gh authentication with user scope for list changes
  • Repository metadata sent to TypeSafe

Getting started

uv sync, export TYPESAFE_API_KEY, gh auth status, then uv run python src/main.py --limit 20 (add --apply to make changes).

At a glance

Type
Project
Status
prototype
Language
Python
Runtime
CLI script
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

Will it change my lists immediately?

No, it is dry-run unless you pass --apply.

What is the default threshold?

0.50; results below are left untouched.

Can I compare with an LLM?

Yes, --compare logs an LLM's answers beside Jev's.

Related projects

  • SiYuan

    Privacy-first personal knowledge management system, listed as integrating Jev for document classification and tag routing.

    PluginproductionTypeScript
  • jev-sheets

    Google Sheets custom functions JEV_IF, JEV_PROB, JEV_CHOICE and JEV_SCORE that classify, tag and score cells with Jev.

    ProjectbetaJavaScript
  • jev-notion

    Native macOS menu-bar app and CLI that fills Notion database properties with Jev judgments, sending uncertain rows to review.

    DemobetaSwift
  • inbox-zero

    Open-source AI email assistant; listed as using Jev as an optional classifier for email intent and triage.

    ProjectproductionTypeScript
  • jev-tab-order

    Chrome extension that groups and orders all tabs in a window by meaning and your rules with a single Jev request.

    ServiceproductionTypeScript
  • orchestkit

    Large Claude Code plugin of skills, agents and hooks that can optionally ask Jev to classify a session's work category.

    ProjectproductionTypeScript