Runtime & Integrations · Runtime Harness
CLI
The entry point — and how it picks an exit code.
The CLI (
argparse) has two commands: run and baseline update.
run composes load config → run → persist → report → render → exit, mapping
the core verdict to an exit class. It never fabricates a verdict or an exit.
One local and CI entry point, with a single job at the end: translate the core’s
ci_should_fail into an exit code. It lazy-imports adapters to avoid an import cycle,
and turns filesystem failures into setup/infrastructure errors rather than quality verdicts.
Commands
bash
# Run an evaluation
evalglass run --config evals/evalglass.yaml [--out reports/] [--format terminal|ci]
# Promote a run record as a baseline (explicit, separate from run)
evalglass baseline update --from reports/run/runrecord.json --to baselines/main.json
Exit codes: 0 for informational /
pass, 1 for fail /
blocked, 2 for a setup or infrastructure error.
It maps, it does not decide
The exit class is a fixed lookup over the verdict the core already produced. The CLI uses the
infrastructure-error class directly only on its own setup/IO path — where there is no Scorecard to
read.