Problem
The current retry prompt sends the full test failure output to the LLM, which is noisy and wastes tokens.
Solution
Parse the pytest failure output into a structured format before sending to the LLM:
{
"failed_tests": ["test_auth_header", "test_route_404"],
"error_types": ["AttributeError: 'Request' has no attribute 'headers'"],
"likely_cause": "FastAPI request object access pattern differs from Flask"
}
This gives the LLM cleaner signal and should improve retry success rate.
Problem
The current retry prompt sends the full test failure output to the LLM, which is noisy and wastes tokens.
Solution
Parse the pytest failure output into a structured format before sending to the LLM:
{ "failed_tests": ["test_auth_header", "test_route_404"], "error_types": ["AttributeError: 'Request' has no attribute 'headers'"], "likely_cause": "FastAPI request object access pattern differs from Flask" }This gives the LLM cleaner signal and should improve retry success rate.