Agent-to-agent handoffs with explicit task ownership

“Research agent, please pass this to verification” is incomplete. A handoff needs one current owner, bounded task, evidence inputs and acceptance criteria.

Validate a handoff record

The MCP contract lab creates an authored transfer.

from mcp_cases import handoff_case

result = handoff_case()
assert result["complete"] is True
assert result["single_owner"] is True
assert result["write_authority"] is False
assert result["handoff"]["status"] == "accepted"

Verification owns a task to label three cited claims using versioned inputs. It has read authority only. No agents run.

Make acceptance explicit

Include task/run ID, previous/new owner, objective, input artifact identities, expected output/schema, acceptance tests, deadline, budgets and authority. The receiver explicitly accepts or rejects; ownership should not be ambiguous.

Transfer durable state and source links rather than a lossy narrative. Do not expand permissions during handoff. Propagate cancellation and trace parent/child IDs.

Test failure paths

Simulate rejected, expired and partial handoffs; missing artifacts; schema mismatch; and duplicated ownership. Evaluate handoff loss against a single-agent baseline.

The Generative & Agentic AI course connects handoff contracts to agent protocols and trajectory evidence.

Exercise

Implement two local workers and a typed handoff. Drop one constraint intentionally, show the acceptance test failing, then repair and resume without duplicate work.

Continue learning

This article is part of the MCP and integration contracts sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in MCP versus a direct API integration: evaluate the trade-off.
• Continue with Detect schema drift in a third-party integration.

Reference: Google A2A protocol documentation.