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

pbi-orchestration-jev

Claude Code skill that splits work into PBIs for Codex, Antigravity or Claude sub-agents, using Jev for model choice and triage.

Projectprototype
View on GitHub ↗

Overview

pbi-orchestration-jev is a Claude Code skill that splits multi-file implementation work into Product Backlog Items and hands them to sub-agents: Codex CLI, Claude via the Antigravity CLI, or Claude Code's Agent tool. The main session orchestrates: it writes PBIs, picks a model per PBI, reviews results and sends work back. The skill text is in Japanese, with English and Japanese READMEs.

Jev handles several judgment calls on short prose only, never source code, diffs or credentials: reading a brief for novelty, coupling and ambiguity to feed a decision table that picks haiku, sonnet or opus; warning about PBI quality; triaging sub-agent reports; and estimating the size of a fix. Pass/fail always comes from running verification commands. Runners are tried in order when one is out of quota, rework is capped at two rounds, and a stats script reports tokens per model and cache rebuilds.

What the decision model decides

TypeQuestionAnswers
ScoreHow novel, coupled and ambiguous is this brief?—
ChoiceHow big is the needed fix (minor, larger, redesign)?(inferred)—
Yes / noDoes this PBI have quality problems?(inferred)yes · no

How it works

  1. Split the work into PBIs and count facts.
  2. Jev judges the brief; select_model.py picks a tier and route.
  3. Sub-agents implement; Jev triages reports.
  4. Run verification commands; rework up to twice or ask the user.

Strengths

  • Model choice by decision table, not gut feel.
  • Only prose sent to Jev.
  • Verification commands decide pass/fail.

Caveats

  • Skill text in Japanese.
  • Requires several CLIs and a TypeSafe key.

At a glance

Type
Project
Status
prototype
Language
Python
Runtime
Claude Code skill
Has code
yes

Use cases and integrations

How curated lists describe it

  • PBI に分けてサブエージェント (Codex / Antigravity / Claude Code) に振り分け、Jev で点検する Claude…

    wh000wh000/awesome-jev-live

Frequently asked questions

What does Jev decide?

Brief difficulty for model selection, PBI quality warnings, report triage and fix size.

Does Jev see my code?

No, only prose such as PBI text, briefs and report excerpts.

Who decides pass or fail?

The orchestrator, by running verification commands itself.

Related projects

  • Jev-AI-Skill

    One skill and MCP server for Claude Code, Codex and Hermes where Jev gates large-model turns, routes models and guards Git steps.

    MCP serverbetaPython
  • jev-skills (n23eos)

    Fourteen opt-in Claude Code and Codex skills where Jev makes one bounded choice: which skill, model, test, file or next step.

    ProjectbetaPython
  • jev-judgment

    Agent skill that sends closed coding-agent judgments to Jev before asking the user, before risky commands and after failures.

    ModelbetaPython
  • jev-opus

    Runs Claude Code on Opus 5.5 with Jev re-picking reasoning effort after every tool batch, via per-message effort so the cache survives.

    ProxybetaTypeScript
  • typesafe-mod

    Claude Code function hook that sends the agent's this-or-that questions to Jev and optionally ranks installed skills per prompt.

    ProjectbetaTypeScript
  • jev-codex-factory

    Multi-agent Codex factory: Jev decides task shape, decomposition and routing, and failures are classified, not blindly retried.

    ProjectprototypePython