Wayflow remote tool rejects valid multiple-output response as a string #226

Description

Wayflow fails the multiple-output remote-tool execution even though the remote endpoint returns a valid JSON object matching the Agent Spec output schema.

The failure is observed in:

• tests/remotetools/validation/test_valid_configs_agent_with_1_remotetool_multiple_outputs.py
• test_valid_configuration_agent_with_1_remotetool_multiple_outputs_can_be_executed

The remote tool is configured with two outputs:

outputs:
- title: sum
  type: number
- title: product
  type: number

The test invokes sum_multiply_tool with a=3 and b=6. The CTS test server returns:

{"sum": 9, "product": 18}

This is the expected response for the declared output schema.

However, the Wayflow log reports:

Expected multiple outputs in a dictionary as result of tool sum_multiply_tool,
but an object of type <class 'str'> was returned.

This indicates that the response is likely being converted from a dictionary:

{"sum": 9, "product": 18}

to a string:

'{"sum": 9, "product": 18}'

somewhere in Wayflow’s remote-tool execution or output-mapping path. Wayflow then validates the converted value, sees a str instead of a dict, and rejects the multiple-output result.

The failure occurs before the expected values can be returned to the agent. The final agent message contains the validation error instead of 9 and 18.

The Wayflow CTS wrapper does not appear to perform this conversion. It passes WayflowFinishedStatus.output_values directly into the CTS result:

AgentSpecFinishedExecutionStatus(
    outputs=execution_status.output_values,
    agent_messages=last_agent_messages,
)

at wayflowruntime/runtime.py:126-129.

Therefore, the likely defect is in Wayflow’s RemoteTool response deserialization, result normalization, or output-schema mapping—not in the CTS assertion itself.

Expected behavior

For the endpoint response:

{"sum": 9, "product": 18}

Wayflow should preserve the response as a dictionary:

{
    "sum": 9,
    "product": 18
}

It should then map the values to the declared Agent Spec outputs and allow the agent to return both 9 and 18.

If the endpoint returns an invalid response, Wayflow should report:

• Remote tool name.
• HTTP status.
• Response content type.
• Response body or a safely truncated response.
• Expected output schema.
• Actual received data type.
• Whether the value was parsed as JSON or retained as text.

Relevant code

• agent-spec/conformance_test_suite/tests/common_server.py:27-35
• agent-spec/conformance_test_suite/tests/remotetools/valid_configs/agent_with_1_remotetool_multiple_outputs.yaml
• agent-spec/conformance_test_suite/tests/remotetools/validation/test_valid_configs_agent_with_1_remotetool_multiple_outputs.py
• agent-spec/conformance_test_suite/runtimes/wayflowruntime/src/wayflowruntime/runtime.py:69-75
• agent-spec/conformance_test_suite/runtimes/wayflowruntime/src/wayflowruntime/runtime.py:118-135
• agent-spec/pyagentspec/src/pyagentspec/adapters/_tools_common.py:33-92