Long-running TypeScript projects accumulate technical debt as 'noise': unused exports, orphaned files, and unnecessary dependencies. Manual cleanup is risky and tedious. A new workflow combines the static analysis tool Knip with an AI Agent to automate the process safely, targeting production code without breaking tests.

The core challenge is distinguishing truly unused code from exports only needed for testing. Deleting the latter breaks test suites. The solution uses a JSDoc `@internal` tag to mark test-only exports. This signals to Knip and future developers to ignore these items during cleanup, keeping the test suite green while streamlining the codebase.

The workflow involves configuring Knip to recognize the `@internal` pattern and then using an AI agent to verify unused code before removal or tagging. This reduces manual verification work. For teams managing large codebases, this approach offers a structured method to reduce noise without introducing regressions.