What problem does it solve?
Optimizing a rich-text editor's performance often devolves into guesswork and unverified claims. This Skill turns Slate v2 performance work into a measured loop: it manages a target registry of benchmarks, enforces correctness gates so speed wins never break editor behavior, and defines stop rules so optimization continues until parity, plateau, or a real blocker.
Core Features & Use Cases
- Target Registry Management: Lists, checks, reports on, and dry-runs benchmark targets defined in
benchmarks/targets/slate-v2.json, and bootstraps new first-class targets when a specific target id is missing.
- Fastest-Safe Optimization Loop: Delegates packet execution to the Slate Autoresearch state machine while applying perf policy, stopping only at target parity, a three-packet plateau under 5% gain, or a correctness/architecture blocker.
- Exactness Gates and Default Contracts: Ships ready-made contracts for pagination/virtualization and huge-document select-all, including cohorts, primary metrics like typing p95 latency, and correctness checks for selection, IME, copy/paste, and undo.
- Use Case: A developer asks to make virtualized pagination fastest for an 800-row document. The Skill dry-runs the pagination target, runs benchmark packets printing
METRIC lines, discards any packet that breaks selection or typing order, and reports baseline, best, and next recommended packet.
Quick Start
Ask the agent to make the Slate v2 pagination route fastest using the virtualized strategy with 800 rows and report the benchmark results.
