Curry Maps 467 Web Screens with Claude Code, Adding 114 New Tests

• Extracted 467 routes from the code to create a screen map, and finalized 104 automation candidates after verification by the responsible parties
• Implemented 114 new tests in 3.5 days, reducing the time by approximately 23 business days compared to manual work, increasing the total number of tests to 578
• Used fault injection testing and AI review skills to prevent judgment errors in advance and discovered multiple actual service defects involving empty banners
• Generated a screen list CSV, mind map, and review sheet to provide evidence, establishing a structure that allows the map to be updated by checking only quarterly code changes

Process and Verification

• Collected routes from three front-end repositories and divided the scope into four sub-agents to identify entry paths through code backtracking, crawler measurements, and existing automation
• Attached evidence such as measurements, code criteria, conditional entry, and automation verification to the paths found by the AI so that responsible parties could review them, while excluding 319 screens with conditional exposure or no entry points

Quality Assurance Strategy

• Injected up to approximately 30 faults, such as hiding elements or delaying responses, to verify failure conditions and ensure tests do not always pass
• Used Claude Code's review skill to have three agents perform rule, risk, and measurement verification, setting a requirement that major changes must pass without issues for 9 consecutive runs