SafePatch AI

Software teams often need to fix bugs under time pressure. A quick patch may solve the visible symptom while introducing hidden regressions, weakening authorization, or creating new security risks. Small engineering teams especially need a reliable way to understand the full impact of a change before it reaches production. SafePatch AI is an evidence-grounded release-safety assistant that analyzes a software repository together with a reported bug, error log, or penetration-test finding. It identifies the probable root cause, connects the conclusion to repository evidence such as file paths and line references, maps affected components, and assigns a transparent severity and confidence level. The application includes a built-in multi-tenant document-management example with a cross-tenant IDOR vulnerability. In Judge Demo Mode, the user can load the sample repository and automatically view the complete workflow. SafePatch AI analyzes route definitions and controller logic, identifies the authorization failure, highlights the affected code, and classifies the risk using security references such as CWE-639 and OWASP API risks. The system then creates a two-level repair plan and generates targeted regression tests for both unauthorized and authorized behavior. A Behavior Comparison view separates the original vulnerable state, the proposed code change, and the expected protected behavior. Proposed changes are clearly labeled and are not silently applied. SafePatch AI is designed around responsible AI principles. It does not execute arbitrary uploaded code, does not invent evidence, and does not commit changes autonomously. A human reviewer must inspect the findings and approve the repair plan before the release-safety report can be downloaded. The target users are developers, technical leads, security engineers, and small engineering teams that need faster but safer software maintenance. The project combines repository-aware development, structured AI analysis, security review, regression-test generation, measurable proof of value, and human approval in one practical workflow.