Where runs come from
Executions lists the runs of your scheduled scenarios and their attempts. The checks that establish whether a workflow is ready aren’t listed here: open them from the workflow on Workflows, under Inspect readiness and history.Find a run
Above the list, narrow by agent and model (All agents and models), repository (All repositories) and status (All statuses). The running and waiting counts at the right are runs in progress; click one to list those runs. Each row shows the prompt, the agent, and when it ran under Ran. Click a row to open the run, or expand it with the arrow at its left to see the model, tokens, duration and status. A run that did not simply finish shows one of these statuses:
The Failed runs card above the list counts runs whose status is Failed. Show failed runs lists exactly those. Timed out, couldn’t start and interrupted runs are not counted there; filter by their status to see them.
Read a run
Click a run to open it. The top shows the agent’s session, its model, the prompt it ran (with Copy prompt), and Repo, Commit, Time, Duration, Tokens and Cost. If the run has more than one attempt, choose one from Attempt; each keeps its own time, spend, record and Results. An earlier attempt never borrows a later attempt’s judgments. While a run is going, Progress shows where it is, and the log so far refreshes every 10 seconds. Once it ends:- Results: workflow completion, the agent’s claim, product choices, workflow steps, success signals and problems supported by that attempt’s record. Open log step takes you to the cited evidence in Logs. Supporting observations expands the remaining observations. Refresh results reads the latest judgments. Not judged yet, judging failed and earlier versions are labelled separately; none is a score or an assumed success. A warning above the results identifies failed evaluations and explains that their outcome remains unknown.
- Above the tabs, the judged outcome sits beside Workflow steps reached. Steps count how far the agent got; the outcome says whether the job was done. Job not done · success check not met means the agent reached the job, but the deciding check did not show it done.
- Problems exposed lists each step your product refused, with its status and a short summary of the answer: a page’s title, an error’s message, or one short line. The answer itself is never shown on the page; Open log step opens the step. When the agent asked one of your docs or website pages for an address it made up, one that nothing it had read, been told or found in a search gave it, the not-found answer is listed as Not a problem, and says so.
- Journey: the agent’s steps grouped into stages, with its existing replay and inspection controls.
- Logs: what the agent did, step by step. focus shows messages, reasoning and tool use; all shows every event. Turn on net to see each request, connection and lookup among the agent’s steps.
- Files: every file the agent added, changed, deleted or renamed, with the change itself.
- Network: a scrollable table with a persistent header shows every request, connection and lookup the run made: the method, URL and status of each request, decrypted, with whether the test key was placed, and the phase each happened in: setup, the agent, or the checks after it.
- Coding agent web activity: the searches it ran and the pages it read.
Repositories
Repositories, under Setup in the sidebar, lists the starter repositories a run can start from. Each is prepared ahead of time, with its setup already done, so a run starts from its code at once. The list is the same for every workspace. Filter it by stack, or sort it. Above the list, the page says how many are Ready.
An Admin’s menu also offers Set as default and Re-run setup. Those stay with the catalogue: the page says the current default is unchanged, and setup is not started again.
Test access
Open Setup → Access to manage test credentials for your product’s surfaces. An organization Admin can choose Add access or Replace access for an API, web-app or MCP server surface. API surfaces take an API key; web apps take a Login and password with a Sign-in URL. An MCP server takes either: choose Login and password for the account your server’s own sign-in accepts, with the server’s address as the Sign-in URL, or API key for a key it accepts as a bearer token. Otherwise the surface determines the kind, so replacing an API key does not offer a login instead. Surfaces come from Brand Vault. Add a surface there first if the list is empty. The list shows each surface and host, its credential type, and whether access is Saved or Not added. Saved means stored; it does not mean a sign-in has been verified. Not added does not mean all runs are blocked. A missing host says Host not set and links to Brand Vault for review. The Account column appears only when a saved credential has an account. For an API or web app, the access panel shows the fixed credential type as text above the fields; for an MCP server it offers Credential type to choose one. Older credentials for other targets remain visible under Manage access, where you can read their account and sign-in page when present and choose Remove access. They are not offered as new test-access targets. Saved secrets cannot be read back on this page. Assignments that use an API key or a web-app test login read its saved value; assignments already issued keep the access they received. Choose Remove access and confirm to stop supplying a credential to future assignments. Members see an explanation asking an Admin to manage access. If a write has no confirmed answer, the page refreshes the list and pauses further writes to that surface for the session. The list cannot confirm which secret was saved. Check the outcome before attempting another write.Who can do what
See People and roles.
The Journey shows the selected attempt’s Judged workflow outcome alongside its recorded activity. Recorded check commands passed counts command results; it does not establish workflow completion. Open Results for the judgment and its evidence, or Logs for the recorded steps.