Workspace Manual Runs and Run History
How one-off manual runs differ from reusable rules, what queued or failed runs mean, and how to read the Runs tab when automation results look stuck or incomplete.
Where you see this in the app
This page documents the Runs tab inside Workspace config & automations.
The tab is split into two user-facing areas:
| Area | What it does |
|---|---|
Manual run | Queue a one-off Codex run for this post |
Recent runs | Review queued, running, succeeded, failed, or canceled runs |
Manual run vs rule
The product intentionally separates reusable automation rules from one-off runs.
| Surface | Best mental model |
|---|---|
Manual run | Do one task now |
Rules | Reuse this trigger automatically later |
If you are testing an idea, recovering from a failed result, or doing a one-time content operation, the Runs tab is the right surface.
Queue run and advanced overrides
The manual-run form is a lightweight launch surface.
| Field | End-user meaning |
|---|---|
Initial prompt | The instructions for this one run |
Workload profile (optional) | Override the default runtime tier for this run |
Advanced overrides | Temporary env/secrets changes for this run only |
Queue run | Submit the run to be processed |
These overrides are intentionally scoped to one run. They do not rewrite the post’s saved defaults.
Recent runs, statuses, and session links
Each run row tells you both the automation state and the launch source.
| Run detail | What it means |
|---|---|
QUEUED | Accepted but not yet being processed |
CLAIMED | A worker picked it up |
RUNNING | The run is actively executing |
SUCCEEDED | The run finished successfully |
FAILED | The run finished with an error |
CANCELED | The run was stopped or invalidated before successful completion |
| session id | The workspace session tied to that run when one exists |
| result summary | A compact output or summary from the run |
| error box | The visible failure reason when one is available |
The trigger label also matters because it tells you whether the run came from a manual action or from an automated event.
How to think about empty, stuck, or failed runs
The user-facing troubleshooting model is:
- no runs: nothing has been queued yet,
- queued for a long time: the background worker may not be processing,
- failed: inspect the visible error and the prompt/defaults that launched the run,
- succeeded with an unexpected result: read the result summary, then adjust the next prompt or defaults.
This tab is a history and diagnostics surface, not only a launch button. It tells you whether the automation system is idle, active, blocked, or producing usable output.
Related docs
See it in action
Previous
Workspace Usage, Credits, and Rolling Windows
How the workspace usage tab reports Codex request counts and token totals, how those numbers relate to your AI credit balance, and what rolling windows tell you about recent activity.
Next
Workspace History, Snapshots, and Rollback
How the History tab works, what Save snapshot creates, what Restore changes, and how users should think about revisions, rollback, and persistence.