Betterbest — Development skill for Claude Code

betterbest

**You've got good. Go for best.** Polish and improve just about anything, from code to prose.

A [Claude Code](https://www.claude.com/product/claude-code) skill for making an existing thing as good as it can be through **iterative experiments** — a prompt, a config, a function, an agent system, an essay, a recipe, a policy doc, a design. Anything you can inspect, change, and measure.

It's a general-purpose adaptation of LLM-driven autoresearch: you improve a *thing* by running one experiment at a time. Change one thing, run a probe, **read** the result against a rubric, keep it if better or revert if not. Repeat until the thing is as good as it can be. The judging is always the model reading the outputs against the rubric — never a regex.

The skill only refines what already exists; it doesn't generate from nothing. You bring the *good*.

What's in the loop

Piece	What it is
Artefact	the thing you're improving (inspectable, mutable, versionable)
Probe	whatever exercises the artefact and yields something to read (test questions at a prompt, a critic-persona panel for prose, tests+profile for code)
Rubric	the weighted dimensions you read each result against — each tagged a hard gate or a soft penalty
Journal	the running record of every experiment and verdict

Plus the disciplines that make it hold up over a long run: judge by reading (never regex), a canary that catches stale-variant runs, a fresh-subagent critique of the judge to fight drift, a co-evolving rubric, and a sealed holdout test run once at the end. See [`SKILL.md`](./SKILL.md) for the full method.

Install

**As a plugin (recommended):** in Claude Code, add the marketplace and install:

/plugin marketplace add drewmccormack/betterbest
/plugin install betterbest@betterbest

**Or as a plain skill:** clone into your Claude Code skills directory:

git clone https://github.com/drewmccormack/betterbest.git ~/.claude/ski