Troubleshooting

I can’t reach the Portal

Confirm that you are using the Portal URL and authentication method provided for your deployment.
Check your browser’s network errors to confirm that the deployment URL is reachable.
Still stuck? Capture the error from the browser DevTools Network tab and contact your Scope administrator.

My request is stuck in `pending`

A request stays pending while it’s waiting for worker capacity.

Bump the priority (range −10 to +10). See Prioritizing & pausing requests.
Check the Portal dashboard for capacity issues.
If pending for hours with capacity available, the worker pool may be unhealthy — contact your administrator.

My run is stuck in `processing`

Open the request’s Logs tab. If the agent is still producing output, it’s working — give it time, especially with the VS Code Electron worker.
If logs are silent for an unusually long time, the agent may be blocked on input or stuck in a tool loop. Cancel and re-submit, or pause/resume.
For repeatable hangs, file an issue with your administrator, including the request ID and worker.

Profile creation fails with HTTP 400 — extensions

Extensions are only accepted by the VS Code Copilot coding agents return HTTP 400 if you try to attach extensions.

Either:

Remove extensions from the profile, or
Switch the profile’s workerType to

See Choosing a coding agent and Defining profiles.

Profile creation hangs on save

If you supplied skill paths or extension IDs without versions, MS Scope is resolving them — skill paths to commit hashes, extensions to Marketplace versions. This may take a few seconds. Pre-pin them to skip the resolution step.

Prompt feature extraction returns HTTP 503

The LLM service powering feature detection (and AI-assisted generation) is unavailable or rate-limited. Wait a moment and retry, or re-trigger from the Portal. See Working with prompt features.

My report shows a criterion as failed but the work looks correct

Open the criterion in the report and read the judge’s rationale. The judge may be looking for evidence that’s present in a different form than expected.
Edit the criterion’s text to be more precise, or override its evaluation prompt. See Defining evaluation criteria.
For DAG criteria, check whether a parent criterion failed — children won’t be evaluated in that case.

Two runs on the same request produced different results

This is expected. AI agents are stochastic. To compare profiles fairly, run several attempts and compare aggregate pass rates rather than single runs.

If results vary wildly, double-check that:

The profile version is the same (the identity may have moved to a new version under you).
agentVersion is pinned in both runs.
Skill and extension references are pinned to specific commits/versions.