Documentation

EvalGlass documentation

Operate the loop with your coding agent, and read what it writes.

EvalGlass is an AI quality-control tool that lets your coding agent build and run project-specific checks for your AI application. These docs are for operating it and reading its results — for humans and for coding agents alike. New here? Start with the Overview, then run the Quickstart.

Install is two commands in Claude Code — /plugin marketplace add Syntelesis-Lab/evalglass then /plugin install evalglass@syntelesis-lab. The deep architecture reference — trust model, evaluation core, and runtime harness — lives under /docs/reference.

Start here

Overview

The loop, what runs where, the four outcomes, the bounded claim.

Start here

Quickstart

One zero-install command to a first, honest Scorecard.

Start here

Agent commands

What to say to your coding agent — the starter prompts and every /evalglass verb.

Start here

Install

Adopt EvalGlass in your own repo, then run.

Run and read

Run an evaluation

Produce a Scorecard and RunRecord locally — no network, no keys.

Run and read

Read a Scorecard

Verdict, authority, status counts, baseline — and what each can’t prove.

Build checks & ship

Build checks

Config

Every evalglass.yaml key and its authority effect.

Ship

Wire CI

Let the verdict drive the build — exit codes, not report prose.

Reference

CLI Reference

Every command, flag, and exit code.

Reference

Scorecard JSON

The primary summary artifact, field by field.

Reference

Codex Runtime

One skills tree, two runtimes.

Going deeper Beyond these product pages, the architecture reference documents the trust model, evaluation core, runtime harness, and vendoring in depth. Every old /reference/… URL redirects to its new /docs home, so existing bookmarks keep working.