What problem does it solve?
Performance tuning often relies on guesswork, leading to speculative optimizations that waste effort or introduce bugs. This Skill enforces a measure-first discipline: capture baselines in the running extension, classify findings by evidence tier, dispatch fixes only for measured regressions or convention violations, and re-measure to confirm real improvement.
Core Features & Use Cases
- Live measurement across four categories: render/webview timing (Lit updateComplete, PerformanceObserver), RPC notification counts and payload sizes, hot-path code audits (missing @memoize, serialized awaits), and git call frequency analysis.
- Three-tier finding discipline: Measured findings require baseline and post-fix numbers, Convention findings fix known GitLens patterns, and Speculation items are filed as open questions and never dispatched to agents.
- Convergence loop with verification: fix agents must re-measure after changes, every agent diff is reviewed, and teardown/relaunch prevents stale-state contamination between baseline and post-fix runs.
- Use Case: A developer notices the Commit Graph feels slow after a branch switch. The skill launches VS Code via the vscode-inspector MCP, captures 5-run averages of hydration time, RPC notifications, and git calls, identifies a missing @memoize on a hot getter, dispatches a fix agent, and confirms the improvement with fresh measurements.
Quick Start
Ask the assistant to run /live-perf on a specific feature, lifecycle stage, or branch diff, for example: measure and improve the performance of the Home webview hydration in the running extension.
