Releases · sadasant/aesthetics

The first release of Aesthetics: evaluate code, documentation, and changes against the qualities you care about.

Install

curl -fsSL https://raw.githubusercontent.com/sadasant/aesthetics/main/install.sh | sh

Or go install github.com/sadasant/aesthetics/cmd/aesthetics@v0.1.0, or download a binary below for macOS, Linux, or Windows (amd64 and arm64).

Commands: check evaluates files, directories, or the whole repository; check --diff evaluates Git changes; explain shows which guidelines apply and why; rules shows configured guidelines and the preset catalog; init writes a starter configuration.
Six presets: unix, kiss, atomic, readme, diataxis, and solid. Each guideline declares its scope, so a preset never judges unrelated files. Known tensions between schools, such as kiss against solid, are declared and resolved by rank.
Your own guidelines, in plain language, scoped by path, file type, content, depth, or artifact kind, and composed with rules, user-defined sets, overrides, and disables.
Outcomes: pass, fail, or indeterminate, derived from calibrated probabilities. Every failure includes a recommendation and, where possible, the lines that show the problem.
Diff attribution: failures are marked introduced or preexisting, and by default only introduced ones fail CI.
Providers: the OpenAI Decisions API, TypeSafe Jev, Cloudflare Clef, self-hosted Laya, any System One endpoint, OpenAI-compatible chat models, or your own command.
CI-ready: non-interactive, with meaningful exit codes, a JSON report, and reproducibility records. It never modifies your files.

Evals on 104 labeled judgments set the defaults:

	OpenAI `gpt-6-luna`	Cloudflare `clef`
Correct on clear cases	97%	98%
False failures	0	0

Read the tutorial to get started, or browse the docs.