-
Notifications
You must be signed in to change notification settings - Fork 1.1k
Description
Description
Im facing below issue while working with Agent-Framework
My Issue:
Currently, we are running these agents in our production environment, which have been developed using the .NET AI Foundry SDK (Persistent Agent). Now, we are trying to use the same agent functionality by using the Microsoft Agent Framework (Sequential Orchestration), but we are encountering JSON parsing issues and are not getting the expected results.
We have modified a couple of agent instructions, which is now allowing the functionality to run correctly.
Question 1:
What is the actual reason behind (root cause) the need to change the instructions of the agents when running through the Agent Framework? Is this because we are using the AI Foundry–Classic agent with the same deployment model, temperature, and top P values?
Earlier, we used the following instructions to invoke the tool in AI Foundry–Classic (Persistent Agent):
Call: myisp_save_solution_parameters(
solutionVersionId, solutionId, enterpriseId,
inputJson (from 2c),
onShoreCountry (from 2b or null),
offShoreCountry (from 2b or null),
nearShoreCountry (from 2b or null),
conversationID (from context),
messageID (from context)
)
The same instructions are not working when invoked through the Agent Framework. Therefore, we updated them as shown below, which resolved the problem. In both cases, we have used the GPT 4.1 deployment model.
** EXECUTE: myisp_save_solution_parameters**
{
"tool": "myisp_save_solution_parameters",
"parameters": {
"solutionVersionId": <int from extraction>,
"solutionId": <int from extraction>,
"enterpriseId": "<string from extraction>",
"inputJson": "<stringified JSON from Step 2c>",
"onShoreCountry": "<string from 2b or omit>",
"offShoreCountry": "<string from 2b or omit>",
"nearShoreCountry": "<string from 2b or omit>"
"conversationId": "<REQUIRED - string from session context>",
"messageId": "<REQUIRED - string from session context>"
}
}Question2:
The input JSON formed in the thread logs for both the failed and successful scenarios is able to invoke the MCP tools successfully using Postman. Why does it fail intermittently only during agent invocation?
postman failure scenerio:
Input Json:
myisp_save_solution_parameters: {"request":{"solutionVersionId":"207475","solutionId":"52869","onShoreCountry":"Usa(100%)","offShoreCountry":null,"enterpriseId":"jagadeesh.madepalli","inputJson":"{"startdate":"09/01/2027","enddate":"08/31/2030","currency":null,"profiles":[{"country":"Usa(100%)","workCountry":"Usa(100%)","payrollCountry":"Usa(100%)","rateType":"GCP | Tier1 | ctyB","economicProfile":"INDUSTRIALIZED TECHNOLOGY SERVICES LONG TERM","city":"Anchorage","isCustomProfile":true,"isDefaultAndCustomProfile":false}],"isDefaultAndCustomProfile":false}","nearShoreCountry":null,"conversationID":"1723cab7-560c-484e-95bd-29cb10e5deec","messageID":"bb9b2e3b-cfbc-4cb8-b427-c65ee5914594"}}
Output:
Postman Success Scenerio:
Input Json:
myisp_save_solution_parameters: {"request":{"solutionVersionId":"207482","solutionId":"52869","onShoreCountry":"Usa(100%)","enterpriseId":"jagadeesh.madepalli","inputJson":"{"startdate":"09/01/2027","enddate":"08/31/2030","currency":null,"profiles":[{"country":"Usa","workCountry":"Usa","payrollCountry":"Usa","rateType":"GCP | Tier1 | ctyB","economicProfile":"INDUSTRIALIZED TECHNOLOGY SERVICES LONG TERM","city":"Anchorage","isCustomProfile":true,"isDefaultAndCustomProfile":false}],"isDefaultAndCustomProfile":false}","conversationID":"1723cab7-560c-484e-95bd-29cb10e5deec","messageID":"afee35f5-a2ad-4462-a136-8e5683250431"}}
Output:
Code Sample
Error Messages / Stack Traces
Package Versions
agent-framework==1.0.0b260116
Python Version
3.13.2
Additional Context
No response
Metadata
Metadata
Assignees
Labels
Type
Projects
Status