Open-source Agent Skills for benchmarking and optimizing Xcode build performance across clean builds, incremental builds, compile hotspots, project settings, and Swift Package Manager overhead.
Install all six skills (the orchestrator needs the specialist skills to work):
npx skills add https://github.com/avdlee/xcode-build-optimization-agent-skillThen open your Xcode project in your AI coding tool and say:
Use the /xcode-build-orchestrator skill to analyze build performance and come up with a plan for improvements.
The agent will benchmark your clean and incremental builds, audit build settings, find compile hotspots, and produce an optimization plan at .build-benchmark/optimization-plan.md. No project files are modified until you explicitly approve changes.
For long-term monitoring across days, machines, Xcode versions, and teams, use RocketSim Build Insights and Team Build Insights.
A 1-second improvement on a 30-second incremental build sounds small. At 50 builds a day, that adds up to 3.5 hours per developer per year -- or 35 hours across a team of ten.
Most projects have several seconds of easy wins hiding in build settings, script phases, and compiler flags. This skill finds them.
The orchestrator coordinates five specialist skills in a recommend-first workflow. Nothing is modified until you approve.
flowchart LR
Orchestrator --> Benchmark["Benchmark\n(clean + incremental)"]
Benchmark --> Compilation["Compilation\nAnalyzer"]
Benchmark --> Project["Project\nAnalyzer"]
Benchmark --> SPM["SPM\nAnalyzer"]
Compilation --> Plan["Optimization\nPlan"]
Project --> Plan
SPM --> Plan
Plan --> You{{"You review\n& approve"}}
You --> Fixer["Build\nFixer"]
Fixer --> ReBenchmark["Re-benchmark\n& verify"]
Phase 1 -- Analyze. The orchestrator benchmarks your project, runs the three specialist analyzers, and produces a prioritized optimization plan at .build-benchmark/optimization-plan.md. No project files are modified.
Use the Xcode build orchestrator to analyze build performance and come up with a plan for improvements.
Phase 2 -- Fix. Review the plan, check the approval boxes for the items you want, and ask the agent to apply them. The fixer implements only approved changes and re-benchmarks to verify.
Implement the approved items from the optimization plan at .build-benchmark/optimization-plan.md, then re-benchmark to verify the improvements.
The plan file is your evidence trail -- shareable with teammates, reviewable in PRs, and diffable over time.
The agent runs over 40 individual checks across build settings, project configuration, source code, and package dependencies.
| Check | What the agent looks for | |
|---|---|---|
| Build settings audit | Debug/Release/General settings against best practices (compilation mode, optimization level, eager linking, compilation caching) | Details |
| Script phase analysis | Missing input/output declarations, scripts running unnecessarily, debug/simulator guards | Details |
| Compile hotspot detection | Long type-checks, complex expressions, compiler diagnostic flags | Details |
| Zero-change build overhead | Fixed-cost phases (codesign, validation, scripts) inflating incremental builds | Details |
| Target dependency review | Accuracy, parallelism blockers, monolithic targets | Details |
| Module variant detection | Config drift across targets causing duplicate module builds | Details |
| SPM graph analysis | Plugin overhead, branch pins, package layering, circular dependencies | Details |
| Swift macro impact | Cascading rebuilds, swift-syntax universal builds | Details |
| SwiftUI view decomposition | Monolithic body properties, result builder complexity | Details |
| Asset catalog parallelism | Single-threaded compilation bottleneck, splitting for parallel builds | Details |
| Access control optimization | Missing final, overly broad visibility inflating compiler work |
Details |
| Incremental build diagnostics | Planning Swift module, SwiftEmitModule, Task Backtraces | Details |
Real-world improvements reported by developers who used these skills. Add your own by opening a pull request.
The xcode-build-orchestrator generates your table row at the end of every optimization run, so contributing is a single copy-paste.
Note
A small clean-build increase is normal when enabling compilation caching -- the first cold build populates the cache. Cached clean builds (branch switching, pulling changes) and incremental builds are where the real gains show up.
| App | Clean Build | Incremental Build |
|---|---|---|
| Helm for App Store Connect | 86s → 91s (+5s / within noise) | 70s → 9s (-61s / 87% faster) |
| Stock Analyzer | 41.5s → 33.2s (-8.3s / 20% faster) | 5.3s → 3.6s (-1.7s / 32% faster) |
| Enchanted | 19.4s → 16.6s (-2.8s / 14% faster) | 2.5s → 2.2s (-0.3s / 12% faster) |
| Wikipedia iOS | 48.7s → 46.5s (-2.2s / 5% faster) | 12.9s → 12.2s (-0.7s / 5% faster) |
| Kickstarter iOS | 83.4s → 83.5s (~0s / within noise) | 10.9s → 10.6s (-0.3s / 3% faster) |
- iOS and macOS teams with slow local build loops
- developers investigating a recent build-time regression
- teams that want evidence-backed Xcode build optimization instead of guesswork
- developers who want a reusable Agent Skills package, not a one-off script
| Skill | Purpose |
|---|---|
xcode-build-orchestrator |
End-to-end workflow: benchmark, analyze, prioritize, approve, fix, re-benchmark |
xcode-build-benchmark |
Repeatable clean and incremental build benchmarks with timestamped artifacts |
xcode-compilation-analyzer |
Swift compile hotspot analysis and source-level recommendations |
xcode-project-analyzer |
Build settings, scheme, script phase, and target dependency auditing |
spm-build-analysis |
Package graph, plugin overhead, and module variant review |
xcode-build-fixer |
Apply approved optimization changes and verify with benchmarks |
The orchestrator is the recommended starting point -- it coordinates the other five skills automatically. Install all six skills so the orchestrator can access each specialist.
Install all six skills (required -- the orchestrator depends on the specialist skills):
npx skills add https://github.com/avdlee/xcode-build-optimization-agent-skillTo install a single skill for standalone use, add the --skill flag:
npx skills add https://github.com/avdlee/xcode-build-optimization-agent-skill --skill xcode-project-analyzerAvailable individual skills: xcode-build-benchmark, xcode-compilation-analyzer, xcode-project-analyzer, spm-build-analysis, xcode-build-orchestrator, xcode-build-fixer. Note that the orchestrator requires all other skills to be installed.
- Add the marketplace:
/plugin marketplace add AvdLee/Xcode-Build-Optimization-Agent-Skill- Install the plugin:
/plugin install xcode-build-skills@xcode-build-skillsTo enable for everyone in a repository, add to your project configuration:
{
"enabledPlugins": {
"xcode-build-skills@xcode-build-skills": true
},
"extraKnownMarketplaces": {
"xcode-build-skills": {
"source": {
"source": "github",
"repo": "AvdLee/Xcode-Build-Optimization-Agent-Skill"
}
}
}
}This repository is packaged for Cursor plugin submission, but the marketplace listing is not live yet.
Once approved, you'll be able to install it from the Cursor Marketplace.
This repository includes an agents/openai.yaml manifest. Copy or symlink the skill folders into your Codex skills directory:
cp -R skills/ "$CODEX_HOME/skills/"See Codex skills documentation for details on where to save skills.
- Clone this repository.
- Install or symlink the specific skill folder from
skills/that you want. - Ask your AI coding tool to use the corresponding skill.
Useful docs: Codex Skills | Claude Code Agent Skills | Cursor Skills
Clean builds expose:
- package and module setup cost
- full project graph overhead
- target structure and explicit-module issues
Incremental builds expose:
- edit-loop pain
- run script bottlenecks
- cache invalidation problems
- repeated package-plugin overhead
That distinction is central to this repo and follows both Apple's Xcode guidance and the SwiftLee workflow in Build performance analysis for speeding up Xcode builds.
Each skill bundles its own copies of the scripts, references, and schemas it needs so it works after standalone installation. The canonical copies live at the repo root:
scripts/-- helper scripts for benchmarking, timing-summary parsing, compilation diagnostics, report generation, and recommendation renderingreferences/-- build settings best practices, artifact format, recommendation format, and source citationsschemas/-- JSON schema for benchmark output
When a root-level file changes, the corresponding copies inside each skill that uses it must be updated (see CONTRIBUTING.md).
skills/
xcode-build-benchmark/
SKILL.md
references/
benchmark-artifacts.md
benchmarking-workflow.md
xcode-compilation-analyzer/
SKILL.md
references/
build-optimization-sources.md
code-compilation-checks.md
recommendation-format.md
xcode-project-analyzer/
SKILL.md
references/
build-optimization-sources.md
build-settings-best-practices.md
project-audit-checks.md
recommendation-format.md
spm-build-analysis/
SKILL.md
references/
build-optimization-sources.md
recommendation-format.md
spm-analysis-checks.md
xcode-build-orchestrator/
SKILL.md
references/
benchmark-artifacts.md
build-settings-best-practices.md
orchestration-report-template.md
recommendation-format.md
xcode-build-fixer/
SKILL.md
references/
build-settings-best-practices.md
fix-patterns.md
recommendation-format.md
All checks are grounded in Apple documentation, WWDC sessions, and proven community practices. See OPTIMIZATION-CHECKS.md for the full list of checks with references to each source.
The stored reference summaries live in references/build-optimization-sources.md.
This repo helps you optimize point-in-time build performance with an agent-guided workflow.
RocketSim complements it by monitoring build performance over time:
- automatic clean vs incremental build tracking
- duration trends and percentile metrics
- machine, Xcode, and macOS comparisons
- team-wide visibility without custom build scripts
If you want to catch regressions earlier and see whether your build times are improving over weeks or months, use RocketSim Build Insights after you apply the improvements from this repo.
Contributions are welcome when they keep the repo focused on Xcode build optimization and Agent Skills format quality.
Please read CONTRIBUTING.md for:
- skill-authoring guidance
- repo scope and quality standards
- workflow notes for scripts and README sync
Created by Antoine van der Lee, creator of SwiftLee and RocketSim. The practical build workflow in this repository is informed by the SwiftLee article Build performance analysis for speeding up Xcode builds and ongoing work on RocketSim Build Insights.
This repository is available under the MIT License. See LICENSE for details.