j-rig

name	j-rig
description	Skill Refiner, the eval-guided improvement loop for SKILL.md files. Runs the bootstrap, score, propose, apply, and status cycle as a thin wrapper over the published @intentsolutions/refiner CLI, proposing safe, minimal, bounded SKILL.md edits and accepting an edit only when a held-out eval score strictly improves with no regression on any other case. Ships a 3-layer cost-tiered hook architecture (sinker, line, hook) that gates skill quality at edit time, end of turn, and commit time. Use when improving an existing skill, refining a SKILL.md against measured behavior, bootstrapping an eval set for a skill, or gating skill edits before they ship. Trigger with "/j-rig", "refine this skill", "bootstrap an eval set", "propose a skill edit", "promote the candidate", or "skill refiner status".
allowed-tools	Read, Write, Edit, Glob, Bash(j-rig:*), Bash(git:*), Bash(python:*)
version	0.1.0
author	Jeremy Longshore <[e-mail]>
license	Apache-2.0
compatibility	Designed for Claude Code; the refine subcommands wrap the published @intentsolutions/jrig-cli binary and run anywhere that CLI installs
argument-hint	[refine bootstrap|score|propose|promote|status] <skill-dir>
tags	["skill-refiner","eval-guided","skill-md","meta-tooling","hooks"]

j-rig — Skill Refiner

Overview

The Skill Refiner is the eval-guided improvement loop for SKILL.md files. It analyzes an existing skill against measured behavior, proposes safe, minimal, bounded edits, and accepts an edit only when it strictly improves. It is the second product in the Intent Solutions agent-rig stack:

J-Rig Skill Binary Eval   ->   Skill Refiner   ->   Rollout Gate
      (test)                     (improve)             (ship)

The refiner proposes bounded edits (add, delete, or replace operations on the SKILL.md text) and accepts an edit only if a held-out eval score strictly improves with no regression on any other case. It never rewrites a skill wholesale, and it never lets a skill judge itself; scoring is delegated to the separate j-rig eval harness.

This plugin is a thin wrapper. The refiner logic lives in the published @intentsolutions/refiner package and is exposed through the j-rig refine command group in the @intentsolutions/jrig-cli binary. This skill invokes that CLI; it does not reimplement any refiner logic.

Use this skill when:

• Improving an existing skill whose behavior you can measure.
• Bootstrapping a held-out eval set for a skill so you have something to score against.
• Proposing a bounded SKILL.md edit and checking whether it strictly improves.
• Gating skill edits before they ship (the 3-layer hooks do this automatically).

Do NOT use it to hand-author a brand-new skill from scratch; use /skill-creator for that. The refiner improves skills that already exist and already have measurable behavior.

Prerequisites

The subcommands wrap the published j-rig binary. Install it once:

# Global — gives you the `j-rig` command everywhere
npm install -g @intentsolutions/jrig-cli

# Or per-repo (recommended for CI version-pinning)
pnpm add -D @intentsolutions/jrig-cli   # then: pnpm exec j-rig --help

The refine command group is contributed by @intentsolutions/refiner, which ships as a dependency of the CLI. Model-backed steps (score, propose) require the ANTHROPIC_API_KEY environment variable; the deterministic steps (bootstrap, apply, ) run fully offline.