Resolve the next blocked step

Check helixor --version and helixor agent --help. If agent is not listed, the installed package does not include the agent commands. Use the agent-enabled package supplied by your administrator or start in the browser.

A local sandbox uses its own sign-in service, which must be running on your machine. Start the sandbox, open its sign-in page, then the console. Do not enter a cloud password into a local sandbox. Shared deployments use their own sign-in URL.

An in-place upgrade from an earlier agent-enabled CLI is not supported. A direct package upgrade was observed to remove SDK files while reporting success. Install the new package in a fresh, administrator-supplied environment and keep the working installation until a supported migration is available. Preserve companion state and indexes; recreating an executable environment is not a reason to delete user data.

Local dev and serve commands require the separate agent runtime. The client package alone cannot start the service. Use an administrator-supplied sandbox environment or connect to the shared console. Do not install unrelated packages to silence this diagnostic.

Check that you are on the intended product console, that Build is enabled, and that your account has the required admin access. The standalone local service does not provide every product feature.

Run helixor agent check on the application and fix the named file and line. Common causes include missing tool argument types and inconsistent messaging declarations. Check again before opening a new test turn.

The browser editor cannot write Python code tools. Pull the draft, edit and check it locally, then push it back as a draft. Follow the CLI round trip.

The draft or live application changed after your copy was created. Reopen or pull the current version into a clean location, compare changes, and reapply your edits. Preserve your work before replacing a local copy.

This is expected. Draft testing holds effects without asking a reviewer or creating an approval. Deploy a practice application to test the live held-call path.

Inspect the required fact, context tool, and argument binding. Restore the read source or correct the rule’s required value. The governed effect does not run when the decision cannot be evaluated.

The human acknowledgement checks the role named by the workflow item. An admin still needs that role. Ask the appropriate team member to review the item.

Refresh the approval record. It may already have been handled. Inspect the existing decision and effect instead of submitting the action again.

The call may have performed its effect even though its final state could not be stored. Do not repeat it. Give the run, tool, and effect references to the person who can verify the external result.

No approved browser execution path is available for this plan. Ask the deployment administrator to check the configured runner and current plan permissions. Changing the agent prompt will not enable it.

The pool credential does not match the current tenant, application and user, or the pool is too old to report that scope. Ask the administrator to verify the deployment and scoped credential. Do not substitute another application's token.

The requested job needs a capability this candidate does not support. Its default transport permits credential-free, network-disabled guest jobs. The explicit local demo additionally admits a managed fixture credential. External websites and saved customer login credentials are not enabled by that example.

Do not repeat the browser action: it may already have happened. Ask the operator to inspect the run, the destination system and the pool session or cleanup records. A failed result is not proof that no action occurred. Keep credentials, page content and cookies out of support messages.

Work lists need the product service. Draft tests and the basic standalone service cannot execute them. In a local sandbox, start the installed product service with helixor agent serve --product and follow the work-list example. Workflow promotion still requires a shared deployment.

The pre-call estimate may exceed the remaining cap. Review max_output_tokens, the selected model, and the applicable agent, tenant, or authoring limit. See Usage and budgets.

Read the trigger’s diagnostic. The reviewed release supports manual runs; unattended scheduler, event-source, and agent-step paths are not active simply because invoke: validates. See Triggers and history.

Check the tenant and reporting coverage. Connected applications must report to the catalog; native Agents and the catalog Dashboard cover different populations. Treat “Not reporting” as missing information.