Remove unneeded node modules from production builds to cut serverless cold starts

Executive Summary

Serverless and container deployments often ship large, unused node packages (for example Prisma engines) that increase cold starts and costs. A CI-integrated pruning and dependency-tracing service would remove runtime-unused files and native binaries before deployment. Serverless and function-first deployments are mainstream and sensitive to package size and cold starts; the source bug is a concrete example of a modern workflow failure (Prisma on Azure Functions). Modern CI/CD pipelines and container build tools make it feasible to run deterministic dependency tracing and lightweight runtime simulations at build time. Additionally, native binaries bundled by JS libraries (Prisma engines) and the continued growth of Node.js monorepos mean bundle bloat is common and costly. The frequency of deploys for cloud-native apps amplifies the ROI of automated pruning: every deploy saves transfer time and GB-seconds costs. Build a CI-integrated dependency-tracer and pruner that analyzes actual runtime entry points, tracks native binaries and platform-specific artifacts, and removes files that are never required in production. Evidence from the source shows a concrete gap: a Prisma user on Azure Functions had @prisma/engines left after npm prune, producing a 62MB package folder that likely is unnecessary. By integrating with common CI systems and cloud providers, the product can offer prescriptive pruning rules, per-runtime simulation, and a registry of known native artifacts (eg Prisma engines) to avoid false positives. Over time the service would accumulate a dataset of dependency usage patterns across frameworks and runtimes, forming a data moat that improves heuristics and reduces breakage risk.