Task permissions and troubleshooting
Understand the executor, connection selection, and reasons a task needs attention.
Understand Runs as
Each task uses its executor’s current project role and only the connections selected for that task that are still enabled in the project. The executor needs at least Contributor access; no separate connection grant is required. Requests from Python follow the same rules.
Saving changes to instructions, schedule, selected connections, permission mode, model, or thinking level explicitly transfers future execution authority to the editor. The edit form explains the transfer. Take over transfers it without changing the instructions. Renaming, pausing, resuming, or manually running an unchanged task preserves the executor.
Each dispatched run freezes its executor and settings. Later task edits affect future runs. Permissions are checked again during execution, so an earlier selection cannot preserve access that has since been revoked.
Select only the required connections
Each task uses its executor’s current project role and only the connections selected for that task that are still enabled in the project. The executor needs at least Contributor access; no separate connection grant is required. Requests from Python follow the same rules.
The chat mode controls execution, while your project role controls what you can do. Read only prevents business-tool changes. Ask for approval requires a Manager’s decision. Automatic permits supported changes for Contributors and Managers unless an approval rule requires review. Choosing a mode never changes project access.
You need Manager access to the proposal’s project to approve or reject changes. Contributors can prepare changes and review the details. Approval rechecks the executor’s current project access and the enabled connection before execution.
Resolve a paused task
- Open Scheduled and filter Status to Needs attention.
- Open the task and read the pause explanation and latest history entry.
- Restore the missing access, edit the unavailable setting, or use Take over with an authorized executor.
- Select Resume when the issue is resolved. The next future occurrence starts the recurring schedule again.
| Reason | What to check |
|---|---|
| Executor cannot access the project | Restore workspace membership and at least Contributor project access, or transfer execution to an authorized project manager. |
| Project is archived | Restore the project before resuming. |
| Selected connection is unavailable | Each task uses its executor’s current project role and only the connections selected for that task that are still enabled in the project. The executor needs at least Contributor access; no separate connection grant is required. Requests from Python follow the same rules. |
| Model or thinking level is unavailable | Choose an available model and supported thinking level, then resume. |
| A business change has an uncertain result | Open the run chat and check the affected records in the connected business tool. Confirm what happened before resuming. |
An uncertain write result pauses future execution immediately. The resume confirmation asks you to confirm that you checked the business records. Steward does not automatically repeat that change. If a correction is needed, handle it as a separate explicit request.
When a run appears late or skipped
Scheduling requires a running Steward worker. During service downtime, a task can remain due. When service returns, one delayed occurrence is dispatched and the next future time is selected. If another run is active, that occurrence is recorded as skipped instead.
Approval and input waits count as active runs. Open the waiting chat to continue or stop it. An intentionally paused schedule does not accumulate runs to execute on resume.
Ordinary errors appear in the chat and run history. They do not replay automatically. Use Run now for a deliberate retry after reviewing the error, or wait for the next recurring execution.
If New task or editing controls are absent, check your project role. If creation reports that execution is unavailable, ask your workspace administrator to check the Steward service configuration.