Reference

Every ATLAS command in the AI Studio skill, grouped by what you’re trying to do. You rarely type these yourself: the coding assistant picks and runs them for you. Knowing the names helps you follow what it’s doing and read your reports.

How the commands run

The assistant calls the bundled CLI from your project root: node .agents/skills/aistudio/scripts/aistudio.js <command> …. Documentation and reports shorten that to aistudio <command>. Locally, tests always run this way; the npm run test targets exist for CI/CD pipelines.

Around a workflow or app edit

CommandWhat it does
do-prettify-workflowNormalizes the workflow file after structural edits.
validate-workflowValidates the workflow and reports its Test Impact.
do-save-workflow · do-save-appSaves the remote DRAFT once, when tests need to run against it. Never publishes.

Plan and inspect (read-only)

CommandWhat it does
get-workflow-test-sync-planLists covered, missing, out-of-date, and review scenarios for a workflow, with the planned action for each.
get-workflow-test-scenariosSuggests scenarios: primary path, branch outcomes, empty-data cases.
get-workflow-test-contextGives the assistant the detail it needs to write a good test definition.
get-workflow-conversation-test-contextLists conversation, HUMAN approval, feedback, and WAIT points before conversation tests are written.
get-workflow-test-judge-contextShows which model node owns the final output, so the judge rubric targets the right thing.
list-workflow-testsLists existing workflow tests with tags, data source, evaluation mode, and data readiness.
get-app-test-sync-planLists an app’s panels, missing or stale panel tests, and the backing-workflow coverage they need first.
get-app-test-contextExplains an app’s testable panels, backing workflows, and widget sources.
list-app-testsLists existing panel tests with tags, data readiness, and evaluation modes.

Create, update, and retire tests

CommandWhat it does
do-generate-workflow-testCreates a workflow test from a definition, with a data capture policy and tags.
do-update-workflow-testUpdates an existing test, keeping tags and unrelated assertions.
do-delete-workflow-testDeletes a test you no longer want, such as an optional variant.
do-generate-app-test · do-update-app-testCreates or updates a panel test, preserving its widget contract.
do-sync-app-testsGenerates or updates all panel tests once backing-workflow tests are done, runs them, and writes the app reports.
mark-workflow-test-unable-to-generateRecords that a scenario couldn’t be reached after real attempts, so only that scenario is deferred.
mark-workflow-test-contract-unreachableExcludes a branch that no valid input can ever reach, with the reason.

Test data and masking

CommandWhat it does
do-record-workflow-test · do-record-app-test-dataRuns the scenario live once and stores the returned data in the test.
do-apply-workflow-test-data · do-apply-app-test-dataApplies compacted or generated data to a test.
get-test-data-masking-contextShows which recorded fields can be masked, without showing their values.
do-create-test-data-masking-profileCreates a reusable masking profile.
do-mask-test-dataPreviews (with --dry-run) or applies masking, then validates file replay.

Run

CommandWhat it does
run-workflow-testRuns one workflow test.
run-workflow-testsRuns a workflow suite, or a slice of it by tags or app.
run-app-test · run-app-testsRuns one panel test or all of an app’s panel tests.
run-workflow · run-workflow-chatAd hoc runs for debugging. These are not tests: no replay data, checks, or reports.

Judging and summaries

CommandWhat it does
attach-workflow-test-judge-resultAttaches one local judge result and refreshes the parent reports.
attach-workflow-test-judge-resultsAttaches all pending local judge results for a workflow or run label.
do-attach-app-test-judge-resultsAttaches pending judge results for panel tests and refreshes app reports.
get-workflow-test-final-summary · get-app-test-final-summaryProduces the evidence behind Validation and Insights: counts, report links, Metrics, Optimization.

Models and optimization

CommandWhat it does
list-workflow-model-configurations · get-workflow-model-configurationLists available model configurations, or shows one in detail.
get-workflow-model-override-targetsShows which nodes can take a different model, and why others can’t.
propose-sweep-candidatesWrites a fresh sweep configuration: node-level by default, workflow-wide with --mode global.
run-optimization-sweepCompiles the sweep report once every candidate profile has run and been judged.
compare-workflow-test-runsCompares two labeled runs: improved, regressed, changed, or unchanged per test.
analyze-token-usageSuggests prompt or grounding reductions from a completed test report.
generate-optimization-reportBuilds token, grounding, and orchestration recommendations for a workflow.

Cost controls

CommandWhat it does
prepare-test-ai-unit-analysis · attach-test-ai-unit-analysisExplains what drove a test over its AI-unit budget and attaches the advice. Changes nothing else.
get-economic-regression-planReturns the next step for setting up or checking an Economic Profile, without changing anything.
run-economic-regressionsEvaluates every configured Economic Profile; can fail the pipeline only where the profile’s policy says so.

Workspace

CommandWhat it does
init-app-packageInitializes a new app-package repository.
get-atlas-recommendations-statusChecks whether ATLAS Recommendations are enabled for a workflow or app.

Flags you’ll see most

FlagMeaning
--data-source file | liveReplay recorded data, or call the connected environment.
--evaluation-mode deterministic | hybridExact checks only, or exact checks then judging. Normally left off so each test uses its own setting.
--judge-provider local | remoteJudge in your assistant (default) or in the Fusion environment (used by pipelines).
--workflow-status DRAFT | PUBLISHEDWhich version to run against. Doesn’t publish anything.
--data-capture-policyrecord-now, record-later, model-generated, or baseline-replay.
--tags a,b · --any-tag a,bRun tests with all listed tags, or with any of them.
--run-labelKeeps comparison and sweep reports separate from normal ones.
--model-overrides · --model-configuration-codeRuntime-only model changes for one run. Files aren’t edited.
--cleanup-scratch trueRemoves temporary files after data is applied or judge results are attached.

Part of the ATLAS learning path reference.

Back to the Learning Path for ATLAS