Agents that have been stuck for more than 5 minutes (or 300 simulated seconds) are listed below. You can hand them off to a human.
The monitor starts with seven fixed sample records: two running, two completed, two stuck and one failed. The filter matches IDs or status text and changes only the visible run list. Filtering stuck shows two records while summary and costs still aggregate all seven. The bars use sixty pixels per record. Add Run chooses random status, steps, tokens and duration locally; it does not observe processes or invoke a model. Edit accepts only the four named statuses, Delete removes a local record, and Reset restores the fixture. These interactions cannot establish actual worker concurrency or completed agent tasks. The saved records contain twenty-eight thousand seven hundred fifty total tokens and stored costs totaling eleven point eight cents. The cost tab sums stored fields rather than recomputing them from token counts. Newly added runs instead use the configured two tenths of a cent per thousand tokens, rounded to four decimal places. Applying that formula to the saved total yields five point seven five cents, which differs from the fixture sum. The bars use four thousand pixels per dollar. This discrepancy is not evidence of actual billing or current provider pricing. Average cost divides stored total by every local record, including failed and stuck records. The handoff list selects records whose status already equals stuck. The five-minute page label is misleading: the function neither checks a three-hundred-second threshold nor promotes running records to stuck. A ten-second refresh only rerenders. Handoff to Human sets the chosen local status to completed and updates its timestamp, then alerts; it sends no human notification and proves no task completion. Share URLs omit last activity and reconstruct timestamps from status on reload. Offline Chart is absent: renderCosts throws after updating list and summary, before handoff and share rendering. Preserve and disclose this partial failure in native checks. This video uses saved source only and invokes no backend or publishing.