Each approval-gated tool call gets its own approval record

When an Agno run pauses a second time on an @approval tool, it now writes a new pending approval record that names only that call, with its arguments and requirements, and continue_run waits until an admin resolves it.

Say an agent pays invoices with a pay_invoice tool marked @approval and @tool(requires_confirmation=True). The run pauses on INV-1, an admin approves that record, and your code continues the run by run_id with no requirements, which is the flow where approvals are resolved in the database. INV-1 gets paid, the model calls pay_invoice for INV-2, and the run pauses again.

In earlier releases nothing appeared for the admin at that second pause. Agno skipped creating a record because an earlier call in the run already had an approval ID, and it then paired the new call with the run's latest record, the approved one for INV-1. The next continue paid INV-2 on the strength of the first decision. We reproduced it offline with a scripted model on both releases.

continue 2 (admin has not approved INV-2):
  -> ValueError: Run has unresolved HITL requirements. Provide the `requirements` parameter (or resolve an admin approval first).
     paid=['INV-1']  records=["approved:['INV-1']", "pending:['INV-2']"]
admin approves 1 record
continue 3: COMPLETED  paid=['INV-1', 'INV-2']  records=["approved:['INV-1']", "approved:['INV-2']"]

When a record lists its tool calls, the run-level lookup pairs it only with those calls, so a call with no record of its own blocks the continue. Tools on a team leader get the same fix. For tools on a member agent, the second record used to name the first call and show its arguments, and it names only the new call now. Audit-mode tools no longer block a continue, metadata["approval"] holds the record for the call being run on an agent or team leader, and resending run.requirements no longer runs a call that already has a result.

See the cookbook, and learn more about approvals in the documentation.