xcode-build-optimization-agent-skill: Benchmark, Plan, and Fix Slow Xcode Builds

A 30-second incremental build touched fifty times a day costs real hours — the README's arithmetic puts a 1-second saving at 3.5 hours per developer per year, and 35 across a team of ten. xcode-build-optimization-agent-skill replaces guesswork with measurements: benchmark, analyze, plan, then fix only what you approved.

Why This Skill Matters

The 1,215-star repo by Antoine van der Lee (creator of SwiftLee and RocketSim) ships six skills. xcode-build-orchestrator coordinates the workflow end to end; xcode-build-benchmark produces repeatable clean and incremental benchmarks; xcode-compilation-analyzer, xcode-project-analyzer, and spm-build-analysis handle compile hotspots, build settings and target dependencies, and the Swift Package Manager graph; xcode-build-fixer applies approved changes and re-verifies.

The agent runs what the README counts as over 40 individual checks, documented with sources in OPTIMIZATION-CHECKS.md across categories including build settings audits, script phase analysis, compile hotspot detection, zero-change build overhead, module variant detection, SPM graph analysis, Swift macro impact, SwiftUI view decomposition, and asset catalog parallelism. The workflow is recommend-first: nothing in your project changes until you check the approval boxes. The repo also publishes a community results table — Helm for App Store Connect reports its incremental build dropping from 70s to 9s (87% faster) — contributed as PRs by developers who ran the workflow.

Installation

Install all six skills — the orchestrator depends on the specialists:

npx skills add https://github.com/avdlee/xcode-build-optimization-agent-skill

For Claude Code there is also a plugin marketplace route (/plugin marketplace add AvdLee/Xcode-Build-Optimization-Agent-Skill), and the repo carries an agents/openai.yaml manifest for Codex-style tools. A single specialist can be installed standalone with the --skill flag if that is all you need.

Real Workflow: From Slow Build to Approved Plan

1. Open your Xcode project in your AI coding tool and start the analysis phase:

Use the xcode-build-orchestrator skill to analyze build performance
and come up with a plan for improvements.

2. The orchestrator benchmarks clean and incremental builds, runs the three specialist analyzers, and writes a prioritized plan to .build-benchmark/optimization-plan.md. No project files are modified.

3. Read the plan. Each item carries its check, its evidence from the benchmark, and a priority. Treat the plan file as your evidence trail — the README pitches it as shareable with teammates, reviewable in PRs, and diffable over time.

4. Check the approval boxes for the items you want, then:

Implement the approved items from the optimization plan
at .build-benchmark/optimization-plan.md, then re-benchmark to verify.

5. The fixer applies only the approved changes and re-benchmarks, so the before-and-after numbers in your plan are measured, not estimated.

Tips

• Expect the first clean build to get slower when compilation caching is enabled — the README warns a small clean-build increase is normal while the cache populates; the gains show up in cached clean builds and incremental builds.
• Install all six skills even if you only talk to the orchestrator — it calls the specialists.
• Clean and incremental builds expose different problems: the README's split is package setup and target structure versus edit-loop pain, script bottlenecks, and cache invalidation. Read both columns of your benchmark, not just the headline number.
• For monitoring trends across days, machines, and Xcode versions, the README positions RocketSim Build Insights (the author's product) as the complement to this point-in-time workflow.

When Not to Use This

It is Xcode-specific — other build systems are out of scope. If you want a one-off sanity check rather than the full workflow, install a single specialist (say, xcode-project-analyzer) with the --skill flag instead of all six. And if a recent regression has an obvious cause — a known slow script phase, a dependency you just added — fixing that directly beats running a full benchmark cycle.

See the leaderboard for more skills.