Test: Validate Agent Framework multi-agent implementation #79

Objective

Validate that an Agent Framework multi-agent workflow can be evaluated locally with the workflow bundle and tool-aware dataset format.

Scope

Use a local agent target with type: agent, hosting: local, execution_mode: local, and a callable adapter. This is the best fit for an Agent Framework multi-agent implementation, and the documented workflow scenario uses agent_workflow_baseline plus dataset rows containing input, expected, tool_definitions, and tool_calls. The target model also explicitly supports framework: agent_framework for agent targets.

Tasks

• Run agentops init
• Point the local callable to the multi-agent workflow entry point
• Set framework: agent_framework in the run config
• Use or adapt the workflow smoke dataset
• Execute agentops eval run
• Verify returned tool calls match the dataset structure
• Review evaluator output for task completion, intent resolution, and tool accuracy
• Log defects and missing guidance

Acceptance Criteria

• A local multi-agent workflow can be executed through AgentOps
• Tool-call metadata is captured in the expected format
• The workflow bundle produces actionable results
• Docs or schema gaps for multi-agent setups are identified
• Clear follow-up backlog items are created

Out of Scope

• Distributed runtime validation
• HTTP deployment of the workflow
• Performance tuning of the multi-agent system