home / skills / glittercowboy / taches-cc-resources / macos-apps
This skill guides building native macOS apps with SwiftUI and AppKit, enabling end-to-end CLI workflows from build to ship with verified outcomes.
npx playbooks add skill glittercowboy/taches-cc-resources --skill macos-appsReview the files below or copy the command above to add this skill to your agents.
---
name: build-macos-apps
description: Build professional native macOS apps in Swift with SwiftUI and AppKit. Full lifecycle - build, debug, test, optimize, ship. CLI-only, no Xcode.
---
<essential_principles>
## How We Work
**The user is the product owner. Claude is the developer.**
The user does not write code. The user does not read code. The user describes what they want and judges whether the result is acceptable. Claude implements, verifies, and reports outcomes.
### 1. Prove, Don't Promise
Never say "this should work." Prove it:
```bash
xcodebuild build 2>&1 | xcsift # Build passes
xcodebuild test # Tests pass
open .../App.app # App launches
```
If you didn't run it, you don't know it works.
### 2. Tests for Correctness, Eyes for Quality
| Question | How to Answer |
|----------|---------------|
| Does the logic work? | Write test, see it pass |
| Does it look right? | Launch app, user looks at it |
| Does it feel right? | User uses it |
| Does it crash? | Test + launch |
| Is it fast enough? | Profiler |
Tests verify *correctness*. The user verifies *desirability*.
### 3. Report Outcomes, Not Code
**Bad:** "I refactored DataService to use async/await with weak self capture"
**Good:** "Fixed the memory leak. `leaks` now shows 0 leaks. App tested stable for 5 minutes."
The user doesn't care what you changed. The user cares what's different.
### 4. Small Steps, Always Verified
```
Change → Verify → Report → Next change
```
Never batch up work. Never say "I made several changes." Each change is verified before the next. If something breaks, you know exactly what caused it.
### 5. Ask Before, Not After
Unclear requirement? Ask now.
Multiple valid approaches? Ask which.
Scope creep? Ask if wanted.
Big refactor needed? Ask permission.
Wrong: Build for 30 minutes, then "is this what you wanted?"
Right: "Before I start, does X mean Y or Z?"
### 6. Always Leave It Working
Every stopping point = working state. Tests pass, app launches, changes committed. The user can walk away anytime and come back to something that works.
</essential_principles>
<intake>
**Ask the user:**
What would you like to do?
1. Build a new app
2. Debug an existing app
3. Add a feature
4. Write/run tests
5. Optimize performance
6. Ship/release
7. Something else
**Then read the matching workflow from `workflows/` and follow it.**
</intake>
<routing>
| Response | Workflow |
|----------|----------|
| 1, "new", "create", "build", "start" | `workflows/build-new-app.md` |
| 2, "broken", "fix", "debug", "crash", "bug" | `workflows/debug-app.md` |
| 3, "add", "feature", "implement", "change" | `workflows/add-feature.md` |
| 4, "test", "tests", "TDD", "coverage" | `workflows/write-tests.md` |
| 5, "slow", "optimize", "performance", "fast" | `workflows/optimize-performance.md` |
| 6, "ship", "release", "notarize", "App Store" | `workflows/ship-app.md` |
| 7, other | Clarify, then select workflow or references |
</routing>
<verification_loop>
## After Every Change
```bash
# 1. Does it build?
xcodebuild -scheme AppName build 2>&1 | xcsift
# 2. Do tests pass?
xcodebuild -scheme AppName test
# 3. Does it launch? (if UI changed)
open ./build/Build/Products/Debug/AppName.app
```
Report to the user:
- "Build: ✓"
- "Tests: 12 pass, 0 fail"
- "App launches, ready for you to check [specific thing]"
</verification_loop>
<when_to_test>
## Testing Decision
**Write a test when:**
- Logic that must be correct (calculations, transformations, rules)
- State changes (add, delete, update operations)
- Edge cases that could break (nil, empty, boundaries)
- Bug fix (test reproduces bug, then proves it's fixed)
- Refactoring (tests prove behavior unchanged)
**Skip tests when:**
- Pure UI exploration ("make it blue and see if I like it")
- Rapid prototyping ("just get something on screen")
- Subjective quality ("does this feel right?")
- One-off verification (launch and check manually)
**The principle:** Tests let the user verify correctness without reading code. If the user needs to verify it works, and it's not purely visual, write a test.
</when_to_test>
<reference_index>
## Domain Knowledge
All in `references/`:
**Architecture:** app-architecture, swiftui-patterns, appkit-integration, concurrency-patterns
**Data:** data-persistence, networking
**App Types:** document-apps, shoebox-apps, menu-bar-apps
**System:** system-apis, app-extensions
**Development:** project-scaffolding, cli-workflow, cli-observability, testing-tdd, testing-debugging
**Polish:** design-system, macos-polish, security-code-signing
</reference_index>
<workflows_index>
## Workflows
All in `workflows/`:
| File | Purpose |
|------|---------|
| build-new-app.md | Create new app from scratch |
| debug-app.md | Find and fix bugs |
| add-feature.md | Add to existing app |
| write-tests.md | Write and run tests |
| optimize-performance.md | Profile and speed up |
| ship-app.md | Sign, notarize, distribute |
</workflows_index>
This skill builds professional native macOS apps in Swift using SwiftUI and AppKit, covering the full lifecycle: build, debug, test, optimize, and ship. It is CLI-first and designed for users who describe requirements while the agent implements, verifies, and reports outcomes. The workflow enforces small, verifiable changes and leaves the project in a working state after every step.
You tell the agent what you want to do (create, debug, add a feature, test, optimize, or ship). The agent selects the matching CLI workflow, runs builds and tests, launches the app when appropriate, and reports concrete outcomes (build status, test counts, launch verification). It asks clarifying questions before ambiguous work and follows a strict change→verify→report loop so every commit is a working state.
Do I need Xcode installed?
Yes. The workflow uses xcodebuild and related CLI tools, so Xcode and its command line tools must be installed.
Will you write tests for UI-only polish?
No. UI polish is usually subjective; tests are written for logic, state changes, regressions, and verifiable behavior. Manual launch and visual review cover polish.