Show HN: I made your coding agent 55% faster at front end (www.react-grab.com)

A developer built "React Grab," a small runtime helper that, when you click an element, walks up the React component tree and embeds a readable component stack (component names + file paths and line numbers) into the DOM. Passing that stack to a coding agent (tested with Claude Code / Claude 4.5 Sonnet) removes the agent’s costly, non-deterministic codebase search phase and lets it jump straight to the right file. In a 20-case benchmark on the shadcn/ui Next.js app, average edit time fell from 17.4s to 7.6s (≈55.9% faster), cost dropped ~47% ($0.75 → $0.40), and average tool calls fell 89.8% (4.4 → 0.5). This is significant for the AI/ML tooling community because it tackles a core "intent → prompt → code → UI" bottleneck: search noise and token waste from language-model-driven exploration. By exposing provenance (exact source locations) in O(1) time, React Grab reduces latency, token usage, and nondeterministic variability—especially useful for low-entropy UI tweaks like spacing or layout. Limitations include framework dependence (React runtime), some outlier tests with increased tokens/time, and the need for non-invasive integrations (e.g., Chrome extension) for broader adoption. The idea — surface precise source provenance to LLMs — is broadly applicable and could materially improve the UX, cost, and reliability of LLM-driven code editing.