An audit-first Agent Skill for developers using Xcode. Reclaim development storage without surprising deletions in any AI coding tool that supports the Agent Skills open format. The skill measures cleanup opportunities, explains the cost of each one, reports recoverable GiB with evidence, and waits for your explicit itemized approval before changing anything.
Built by Antoine van der Lee (SwiftLee), known for RocketSim, which comes with a CLI + Agent Skill for Simulator interactions, and his other agent skills like the popular SwiftUI and Swift Concurrency ones.
- Developers whose Mac keeps running out of space and suspect Xcode is the reason
- Anyone with years of accumulated DerivedData, simulators, runtimes, and archives
- Beta testers who install every Xcode beta and end up with leftover platform runtimes
- Teams that want agents to clean up build machines safely, with an audit trail
The skill runs a read-only audit across every place Xcode quietly accumulates storage, from build caches to forgotten beta platform runtimes. Each finding is reported with its size in GiB, supporting evidence, and the cost of regenerating or permanently losing it.
- Xcode DerivedData for missing or stale workspaces
- Project-local
.buildand.derived-datafolders - Available and unavailable Simulator devices
- SwiftUI Preview device sets
- Installed Simulator runtimes, including stale beta runtimes superseded by a newer platform version that no installed Xcode SDK still uses
- Orphaned Simulator runtime volumes not registered with CoreSimulator
- Simulator dyld caches left behind by removed runtimes
- Simulator clones created by parallel test runs (
XCTestDevices) - CoreSimulator and iOS device diagnostic logs
- CocoaPods download caches
- Documentation caches and legacy DocSets
- Archives, dSYMs, and per-platform DeviceSupport symbols (iOS, watchOS, tvOS, visionOS), keeping the newest symbol set per platform
- Orphan archives that were never uploaded or exported and are superseded by a newer archive of the same app
- Xcode-managed components such as the Metal Toolchain and downloadable documentation assets
- Downloaded Xcode
.xip,.dmg, and.zipinstallers - Installed Xcode applications, active selection, and evidence of recent use
As a developer you always need to stay in control of what gets deleted from your machine: a build cache is cheap to lose, but the wrong archive or simulator is not. That's why the skill treats “clean Xcode” as permission to inspect, never as permission to delete.
Before anything is mutated, you get to see for every candidate:
- A stable candidate ID
- Exact path or Simulator identifier
- Measured bytes and GiB
- Risk and regeneration cost
- Evidence supporting the recommendation
- The exact action it proposes
Nothing happens until you approve exact IDs. Ordinary files move to Trash by default, so you can still recover them. Irreversible Simulator operations ask you for a second, separate confirmation. Your archives and dSYMs stay preserved unless you confirm the exact distributed build is safely backed up.
Unlike text-only skills, this repository ships an executable Python script that does the measuring and the (gated) cleanup, so your agent reasons over structured JSON instead of ad-hoc shell output.
When it triggers:
- You mention low disk space, Xcode storage, DerivedData, simulators, runtimes, archives, or old Xcode versions.
- You ask why Xcode Settings → Components still lists old beta platforms.
- Your agent runs into low storage on its own, for example when a build or download fails with an out-of-disk-space error.
What you can ask:
Audit my Xcode-related disk usage and show how many GB each cleanup could recover.My Mac is low on storage and Xcode seems to be the reason. Figure out what's safe to remove.Which simulator runtimes are actually stale after installing the latest Xcode beta?Find Xcode installers in my Downloads folder that I no longer need.
Under the hood:
scripts/xcode_disk_cleanup.py audit: read-only inventory across all categories. Emitsaudit.json(stable candidate IDs, sizes, risk, evidence, path fingerprints) and a markdown report.scripts/xcode_disk_cleanup.py apply: executes only the exact candidate IDs you approved, with a required confirmation phrase, a second phrase for irreversible Simulator operations, path-identity revalidation, and running-process checks before every mutation.- Evidence-based staleness: simulator runtimes are cross-checked against every
installed Xcode with
simctl runtime match list, archives against their Organizer distribution records, DerivedData against recorded workspace paths, and installers against installed Xcode versions.
- SwiftUI Expert
- Swift Concurrency Expert
- Swift Testing Expert
- Core Data Expert
- Xcode Build Optimization
- Xcode Simulator AI Control Agent Skill
Install this skill with a single command:
npx skills add https://github.com/AvdLee/Xcode-Disk-Cleanup-Agent-Skill --skill xcode-disk-cleanupThen use the skill in your AI agent, for example:
Use the xcode disk cleanup skill and show me how many GB I can safely recover.
- Add the marketplace:
/plugin marketplace add AvdLee/Xcode-Disk-Cleanup-Agent-Skill- Install the Skill:
/plugin install xcode-disk-cleanup@xcode-disk-cleanup-agent-skillTo automatically provide this Skill to everyone working in a repository,
configure the repository's .claude/settings.json:
{
"enabledPlugins": {
"xcode-disk-cleanup@xcode-disk-cleanup-agent-skill": true
},
"extraKnownMarketplaces": {
"xcode-disk-cleanup-agent-skill": {
"source": {
"source": "github",
"repo": "AvdLee/Xcode-Disk-Cleanup-Agent-Skill"
}
}
}
}When team members open the project, Claude Code will prompt them to install the Skill.
This repository includes an agents/openai.yaml manifest. Copy or symlink the
xcode-disk-cleanup/ folder into your Codex skills directory:
cp -R xcode-disk-cleanup/ "$CODEX_HOME/skills/xcode-disk-cleanup"See the Codex skills documentation for details on where to save skills.
Install via pi:
pi install https://github.com/AvdLee/Xcode-Disk-Cleanup-Agent-SkillThe skill will be available automatically in pi sessions.
- Clone this repository.
- Install or symlink the
xcode-disk-cleanup/folder following your tool's official skills installation docs (see links below). - Ask your AI tool to use the “xcode-disk-cleanup” skill when storage runs low.
Follow your tool's official documentation, here are a few popular ones:
- Codex: Where to save skills
- Claude: Using Skills
- Cursor: Enabling Skills
How to verify:
Your agent should run the bundled audit script first, present an itemized report with candidate IDs and recoverable GiB, and refuse to delete anything until you approve exact IDs.
xcode-disk-cleanup/
SKILL.md - Safety contract, audit workflow, and proposal format
references/
safety-model.md - Confirmation phrases, Trash-first behavior, protected paths
generated-data.md - DerivedData, project .build folders, CocoaPods, documentation
simulators.md - Devices, runtime staleness evidence, dyld caches, XCTest clones
release-artifacts.md - Archives, distribution records, DeviceSupport, logs, DocSets
xcode-installations.md - Installer matching and multi-signal Xcode usage evidence
scripts/
xcode_disk_cleanup.py - Read-only audit and confirmation-gated apply
tests/ - Deterministic unit tests for classification and apply gating
evals/ - Scenario evaluations used to benchmark the skill
- Token efficient: one bundled script performs the entire audit and returns structured JSON, so your agent reasons over compact data instead of burning context on dozens of exploratory shell commands.
- Proven Apple tooling: cleanup goes through supported workflows like
xcrun simctland Xcode itself, never through wildcard deletes,sudo, SIP changes, or raw system-runtime removal. - Evidence over heuristics: staleness is proven with SDK runtime matching, Organizer distribution records, workspace existence, and installer version matching before anything is proposed.
- Honest accounting: deterministic JSON and Markdown reports separate candidate sizes from actual APFS space recovered.
- Safe by design: read-only by default, Trash first, and path identity plus running-process revalidation right before every mutation.
The safety model prioritizes first-party guidance:
- Managing Xcode components
- Building with debugging information
- Symbolicating crash reports
- Configuring command-line tools
- SwiftPM cleaning builds and caches
- APFS shared space
Contributions are welcome! This repository follows the Agent Skills open format.
Please read CONTRIBUTING.md for format requirements and the pull request process. Cleanup rules should favor evidence and supported APIs over aggressive deletion.
Created by Antoine van der Lee, the creator behind SwiftLee and RocketSim, and a developer since 2009. This skill distills years of Xcode housekeeping experience into evidence-based guidance for AI assistants.
This skill is open-source and available under the MIT License. See LICENSE for details.