Automation and JSON

Use JSON modes in scripts. Human output is for people and may evolve. In Bash or Zsh, enable pipefail so a formatter such as jq does not hide a failed Glade command. Replace <base-ref> with the intended existing comparison ref and verify it resolves in the checkout.

set -o pipefail
glade check --project . --json --no-progress | jq .
glade test changed --project . --since <base-ref> --json --no-progress | jq .

Envelope

Priority commands use a versioned envelope.

{
  "schemaVersion": "1.0",
  "command": "check",
  "status": "failed",
  "exitCode": 1,
  "project": {},
  "summary": {},
  "diagnostics": [],
  "artifacts": [],
  "suggestions": []
}

Fields after exitCode are command-dependent and may be omitted. During migration, command-specific legacy data may appear under data; some commands still use their own JSON shape.

Inspect saved test JSON as well as the process status. The summary contains total, passed, failed, errors, skipped, and unsupported counts. Unsupported test outcomes count as errors and exit with code 1. An empty affected selection can exit with code 0; run an explicit relevant test or suite when execution evidence is required.

excludedClasses names invalid classes, their dependents, and excluded triggers; summary.excludedClasses counts them. Any exclusion exits 1, even when all executed tests passed.

Local Flow execution

Local DML loads active record-triggered Flows from project metadata. The recordTriggerType selects create, update, create-and-update, or delete; local execution handles before-save and after-save Flows, plus explicit Delete Flows in the before-delete phase. After-delete Flows are not executed locally. The checked paths cover entry filters, decisions, assignments, supported record lookup/create/update/delete, loops, subflows, and Apex actions. For supported after-save entry criteria, doesRequireRecordChangedToMeetCriteria compares the current record with its prior state. $Record and $Record__Prior support the captured current/prior record behavior. Invocable.Action can call supported local Flow actions and returns local result/error objects.

Screen and wait nodes remain unsupported. Hosted scheduling, interactive Flow screens, and the full hosted Flow engine remain outside the local contract. An unmodeled node is not proof that the whole Flow runs correctly.

glade check reports GLADEAUTO001 workflow warnings, GLADEAUTO002 Flow warnings, and GLADEAUTO003 for an automation file skipped because it could not be read or parsed. Other files continue to load. GLADEAUTO004 is an error for invalid supported Flow definitions, such as a bad triggering-record field reference. Inspect diagnostics before relying on an automation result.

Stdout and stderr

Stdout carries the primary result: JSON, SARIF, JUnit, Markdown, HTML, or terminal text.

Completed result-producing JSON modes keep progress on stderr, including --progress-json NDJSON events. Early setup failures may leave stdout empty; check process status before parsing. Watch modes emit NDJSON on stdout, while wizard output and legacy paths such as glade test failed --json with no saved failures still print human text. See CLI output modes.

CI pattern

- uses: actions/checkout@v4
  with:
    fetch-depth: 0
- run: test -f glade.yml || glade init --project . --yes
- run: glade doctor --project .
- run: glade check --project . --format sarif --output glade-check.sarif
- run: glade test changed --project . --since <base-ref> --json --no-progress
- run: mkdir -p reports
- run: glade test --project . --junit reports/glade-junit.xml

Use if: always() only for artifact upload steps that must run after a failing check or test command.

Specialized formats

Command	Machine formats
glade check	JSON, SARIF, GitHub annotations
glade test	JSON, JUnit
glade report	JSON, HTML, Markdown, GitHub annotations, zip
glade debug profile	JSON, text, Markdown
glade profile analyze	JSON, text, Markdown, pprof