performance-optimizer_skill

Overview

This skill analyzes and optimizes parallel development workflows to find bottlenecks, estimate time and cost savings, and recommend high-impact fixes. It focuses on setup, execution, and integration phases and provides prioritized, actionable steps to improve parallel efficiency and ROI. Use it to benchmark performance and estimate theoretical vs actual speedups.

How this skill works

The skill collects timing and resource metrics (setup, execution, integration), computes parallel efficiency and Amdahl's Law speedup, and scans plans for sequential patterns or hidden dependencies. It diagnoses common bottlenecks—sequential setup, implicit task dependencies, resource overload, and slow merges—and returns concrete changes, estimated savings, and implementation guidance. It can also produce before/after comparisons and simple benchmarking scripts.

When to use it

• You notice slow parallel execution or long setup times
• You want to know theoretical vs actual speedup (Amdahl's Law)
• You need prioritized optimizations and estimated time/cost savings
• You want benchmarking or continuous monitoring of parallel workflows
• You need ROI or cost comparisons for alternative agent patterns

Best practices

• Measure current state first: capture planning, setup, execution, integration durations
• Eliminate sequential setup: let subagents create their own issues/worktrees in parallel
• Detect and extract hidden dependencies to enable true concurrency
• Limit concurrent agents to match system resources and batch if needed
• Merge incrementally and run tests in parallel or once after merges

Example use cases

• Diagnose a 5-task workflow where setup dominates and convert it to parallel setup pattern to save 30% of setup time
• Find a hidden dependency causing one long-running task and refactor into a small shared interface to cut total time by ~40%
• Recommend batching limits when memory/CPU is saturated to avoid swapping and reduce runtime
• Reduce merge time by switching to incremental merges and feature flags, lowering merge overhead from ~90 minutes to ~20 minutes
• Run a quick benchmark script to quantify setup overhead and average time per task for continuous improvement