A paused run continued in the background no longer repeats its tool call

When an Agno agent or team resumes a human-in-the-loop run with background=True and stream=True, the model now sees the paused tool call once. A background continue marks the run RUNNING before it rebuilds the messages, and history only skipped runs marked PAUSED, CANCELLED, ERROR or REGENERATED. So the paused turn also came back as history, with its tool call but no result, and OpenAI Chat Completions refused the request:

An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'.

The AgentOS UI sends both flags on every continue, so any agent or team with add_history_to_context=True hit this on confirmation, user input and external execution.

Agno now finds the continued run by its id, whatever status it has stored, and leaves it out of history before applying num_history_runs and num_history_messages, so earlier turns still reach the model. Thanks to rodrigocoliveira, who found the root cause.

See the cookbook, and learn more about human-in-the-loop and background execution in the documentation.