AGENTS.md

AGENTS.md

Run the checks locally

The checks need an engine at this SDK's floor, v1.0.0-beta.18, or later. The CLI runs on the engine it ships with; --engine or DAGGER_ENGINE selects another (dagger help engine):

export DAGGER_ENGINE=container://<running-engine>

Run the e2e checks from the root of this checkout, by name or all of them:

dagger check -m .dagger/modules/e2e -l
dagger check -m .dagger/modules/e2e runtime-call-check find-client-root-check
dagger check -m .dagger/modules/e2e 'e2e:*'

Without a pattern, dagger check also selects the engine-e2e checks, which build engines from dagger/dagger commits. CI runs them:

dagger check engine-e2e:runtime-call-check
dagger check engine-e2e:floor-coverage-check
dagger check engine-e2e:below-floor-check

Each e2e check has an engine-e2e check of the same name, which runs it on the floor engine, built from its commit. floor-coverage-check fails when an e2e check has no such check. To run e2e checks on the floor's released engine instead:

dagger check --engine=image://registry.dagger.io/engine:v1.0.0-beta.18 \
  -m .dagger/modules/e2e runtime-call-check

Two engine-e2e checks run the target checks on the engine of the release this SDK targets, v1.0.0-beta.24, built from its commit. Only that release serves a module or a session the view of the release it announces, so the floor cannot prove it. floor-coverage-check does not count these checks:

dagger check engine-e2e:target-version-on-target-check
dagger check engine-e2e:target-version-cli-on-target-check

Move their commit with the targeted release.

Judge a check by dagger check, never by dagger call <check>: on v1.0.0-beta.18, dagger call exits 0 when the check fails.