Claude Code can arm monitor_ws tasks for artifact live updates. T3 does not recognize that task type as a monitor, so the shared task classifier stamps it as an agent.
That makes the Agents panel show a non-model websocket watcher as “Working” with the parent model/effort, and it also affects background-liveness labels.
Add monitor_ws to the existing MONITOR_TASK_TYPES set.
This keeps the change at the canonical classifier already used by ingestion, sidebar liveness, and client fallback. No provider-specific UI handling is added.
The focused contract test now covers monitor, monitor_mcp, and monitor_ws as background work.
Fixes #13978.
Focused regression coverage is included in providerRuntime.test.ts. Local execution was not available from this GitHub-only session, so this is opened as a draft for CI verification.
Model: GPT-5.6 Sol. Harness: ChatGPT.
Added 1 commit
Note
This comment is posted by Julius' dot
The proposed classifier change matches the accepted scope in #13978, but the PR says its regression test was not run, and current checks contain only label jobs. There is no successful check of the changed classification under verification. Please run the focused contract test and report its result, then request reconsideration.
