AI Code Tool Output Review Gate (QA)

AI Code Tool Output Review Gate (QA)

1. 01

   Define acceptance criteria for AI-generated code before review begins

   Specify functional requirements, performance thresholds, and security constraints. For example, require that generated code passes existing unit tests and adheres to the project's linting rules.

2. 02

   Run automated security scans on all AI-generated code

   Use tools like sentinel-scan-cli to test for prompt injection vulnerabilities and other common issues. This provides a concrete audit trail for client-facing deliverables.

3. 03

   Verify that generated code integrates with existing architecture

   Check that the AI tool's output respects the project's design patterns, data flow, and dependency management. This prevents architectural drift and reduces technical debt.

4. 04

   Conduct a manual code review focusing on logic and edge cases

   AI models can miss subtle business rules or fail on unusual inputs. A human reviewer should trace through critical paths and test boundary conditions.

5. 05

   Assess performance and cost implications of the generated solution

   Compare the runtime efficiency and resource usage against the original requirements. For instance, if the AI tool suggests a new algorithm, benchmark it against the current implementation.

6. 06

   Document any modifications made to the AI-generated code

   Keep a record of changes for future reference and to inform the agency's learning. This also helps in maintaining consistency across client projects.

7. 07

   Approve or reject the output based on the established criteria

   If the code fails any critical check, send it back to the AI tool with specific feedback or fix it manually. The goal is to maintain quality without slowing down delivery.