Harmonize skill
A Swift architecture linter with rules as unit tests, and a Claude Code plugin with skills to run and write the rules.
- 364 ★ on the repository
- Last commit
- Apache-2.0
Minimal access
Why this level
- The plugin runs project tests and writes skill files into a chosen repository folder
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/harmonizeDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The plugin runs project tests and writes skill files into a chosen repository folder
This is third-party code. Review the repository files before installing.
What it does
Harmonize checks Swift code structure with tests in Quick, XCTest or Swift Testing, for example that every class ending in `ViewModel` must inherit `BaseViewModel`. A rule is described by a `Rule` type with rationale, fix, bad and good examples, so a violation explains itself. For agents the repository ships a Claude Code plugin with two skills: `run-harmonize` runs the rules after a task and explains the output, and `write-harmonize-rule` guides an agent writing rules. A Swift package command also generates a separate skill with a `SKILL.md` from the rules. The agent integration is in beta, version `2.0.0-beta.2`.
Who it is for. iOS and macOS teams where an AI agent writes code alongside people and architectural rules need to be checked rather than reviewed by hand.
Good fit when
- A Claude Code agent finished a task and the rules must run automatically.
- You want to record an architectural convention as a test, for example about inheritance.
- You want to give the agent a skill built from the rules so it knows conventions before writing code.
- The rules should run in CI beside ordinary tests.
Not a fit when
- You only need code style, for which linters like SwiftLint fit.
- The project is not written in Swift.
- The agent integration is in beta and stability matters more.
Example request
After the edits, run the Harmonize rules with the run-harmonize skill and fix violations using each rule's howToFix field.Limitations
Agent features are in version `2.0.0-beta.2` and the format may change. A Swift project with the package in `Package.swift` is required. To build the skill the plugin writes files to a chosen folder, and the `--allow-writing-to-directory` value must match the `output` setting. Rules are written in Swift and need familiarity with test frameworks, and the choice among Quick, XCTest and Swift Testing is made up front. Before use, check the repository's current state and the version you rely on.
How to disable. Run `/plugin uninstall harmonize@harmonize` and remove the `run-harmonize` line from `AGENTS.md` or `CLAUDE.md`. Take the Swift package out of `Package.swift`.
FAQ
How does a rule differ from a test?
It is written as a test but carries a `Rule` object with rationale, fix and examples that the agent reads.
What do the plugin's two skills do?
`run-harmonize` runs the rules and explains output; `write-harmonize-rule` helps an agent write new rules.
How do I get a skill from my rules?
The Swift package command plugin reads `Rule(...)` declarations and creates a `SKILL.md` with reference files in the folder set in `.harmonize.yaml`.
Can it run in CI?
Yes. The rules run as unit tests, and the README gives a GitHub Actions example.
Which test frameworks are supported?
Quick, XCTest and Swift Testing.
Discussion
No questions or reviews yet.
Related
SkillsSwift Testing Agent Skill
swift-testing-agent-skill
A Swift Testing skill from Antoine van der Lee: the new APIs, migration from XCTest, parameterized tests and flat tests. For Claude Code and other agents
Paul Hudson's Swift Testing skill: parameterized tests, confirmations, exit tests and a move from XCTest to the new framework
Validates Agent Skills packages: structure, links, token weight and instruction quality, with optional LLM-as-judge scoring
The evidence-code-review skill for Codex and Claude Code: evidence-based change review with MUST, SHOULD, BETTER and NITS levels
A Playwright and TypeScript project template with rules and 17 skills so Claude Code, Codex and Cursor write tests to a standard
Two skills for Codex and Claude Code: a read-only repository audit and an experiment loop with saved baselines.
README in short
Architecture rules live as tests, and the agent gets skills to run them, write them and export them as a project skill.