roadmap / ios
CLAUDE.md11 starsChanged 5 days agoMITView on GitHub
Rules for agents working under ios/. Budget: ≤100 lines, one-in-one-out. A new rule replaces or tightens an old one; detail goes to architecture.md or docs/gotchas.md. The root CLAUDE.md applies in full (worktrees, reviews, delegation, stories, security). This file adds only what is specific to the iPhone app. - Every UI change lands on both clients, the iPhone app and the web widget, in the same task (Brad 2026-11-02). Name the iOS twin (ios/HealthRoadmap/...) and the web twin (widget-src/src/components/...). Ship both,…
In this repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- codex-reviewSkill
- design-reviewSkill
- untangleSkill
The file
We split this CLAUDE.md into the rules it gives an agent
8 rules. About 0 are common, 2 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 11
Every decision and its why: architecture.md.
- Very specific: line 17
Every UI change lands on both clients, the iPhone app and the web widget, in the same task (Brad 2026-11-02).
- Very specific: line 31
Keep DerivedData outside ~/Documents (-derivedDataPath <scratch>/dd): the repo sits in an iCloud-synced folder, Finder adds xattrs to build …
- Very specific: line 33
Never name a source file *backup*: .gitignore hides it on this case-blind Mac, and Xcode still builds it, so a fresh clone breaks.
- Very specific: line 49
Run a sim-run command that may wait in the background (run_in_background).
- Very specific: line 59
Only for a demo device, prefix ALLOW_EXTRA_SIMULATOR=other-device.
- Distinctive: line 66
No test data on a real device (Brad 2026-11-02).
- Very specific: line 87
Search docs/gotchas.md for "iOS", "iPhone", "simulator" and "xcodebuild" for the rest.
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
ios/CLAUDE.md
Rules for agents working under ios/. Budget: ≤100 lines, one-in-one-out. A new rule replaces or
tightens an old one; detail goes to architecture.md or docs/gotchas.md.
The root CLAUDE.md applies in full (worktrees, reviews, delegation, stories, security). This file adds only what is specific to the iPhone app.
Where things are decided
- Every decision and its why: architecture.md. Read the section you touch; keep it current.
- Release and store steps: launch-checklist.md, app-store.md.
- Over-the-air engine updates: engine-updates.md.
Both clients, one engine
- Every UI change lands on both clients, the iPhone app and the web widget, in the same task (Brad 2026-11-02). Name the
iOS twin (
ios/HealthRoadmap/...) and the web twin (widget-src/src/components/...). Ship both, or say which side waits and why. When Brad's request names one app's layout, ask before mirroring it. - Shared logic lives once, in health-core, which the app runs as the bundled engine. Never re-implement it in Swift.
- The upload runs through the optional
upload*Jsondoors: serial steps, the view coalesced to one call per frame off the main thread (architecture.md D3, fifth optional door).
Build
- Build the engine first:
node ios/scripts/build-engine.mjs. Its outputs (the engine bundle, itshealth-engine.jsonsidecar,HealthRoadmapTests/Fixtures/) are generated and gitignored. - Silent trap: an Xcode build does not rebuild the engine. After a health-core change, a build from
Xcode runs the old engine until
build-engine.mjsorios/scripts/test.shruns. - Keep DerivedData outside
~/Documents(-derivedDataPath <scratch>/dd): the repo sits in an iCloud-synced folder, Finder adds xattrs to build products, and codesign refuses them. - Never name a source file
*backup*:.gitignorehides it on this case-blind Mac, and Xcode still builds it, so a fresh clone breaks. Check an unlisted new file withgit check-ignore -v <path>.
xcodebuild and simulators (shared by every session on this Mac)
- One simulator for every session (US-71 AC7): "HR Shared iPhone", reached only through
ios/scripts/sim-run.sh -- <command>(xcodebuild test, build-for-testing, test-without-building; simctl device commands;{udid}in the command is the device). It holds a kernel lock (lockf) for the run: a second caller waits, naming the holder (30 min,HR_SIM_WAIT_MIN). It adds the device and-parallel-testing-enabled NO, stops a run after 30 minutes, keeps the device booted, and shuts it down after 20 idle minutes. Check your work and let the command finish; never hold it while coding. --reset-appuninstalls the app and clears its simulator settings, so no data leaks between sessions.- Demos and the iOS Simulator MCP tool (outside the guard) use the demo device, "HR Demo iPhone", never
the shared one:
ALLOW_EXTRA_SIMULATOR=other-device ios/scripts/fresh-sim.shmakes, boots and fills it. Or wrap all the steps in one script run through sim-run. Shut the demo device down when done: sim-run refuses (75) while it is up. - A wait can last 30 minutes, longer than the Bash tool's 10-minute limit, which would kill it.
Run a sim-run command that may wait in the background (
run_in_background); it also waits while the load is 8+ or two Codex reviews run. ios/scripts/sim-run.sh status: both devices, the holder, leftover simulator processes, and the cleanup commands for Brad to run (never kill simulator processes yourself). Its tests need macOS (lockf), so CI skips them: runnpx vitest run ios/scripts/sim-run.test.tsafter changing it.- Any other
xcodebuild(archive, plain build):pgrep -fl xcodebuildfirst; if one runs, report back. Time-limit it:perl -e 'alarm shift; exec @ARGV' 1200 xcodebuild .... A long-open Xcode app is the likely cause of hung builds; quit it. - The guard
.claude/hooks/simulator-guard.cjsdenies simulator use outside sim-run.sh, anxcodebuild testwithout-parallel-testing-enabled NO, andsimctl create|cloneonce two are booted. It is a seat belt, not a wall: it reads one command line, not scripts or every shell trick. Only for a demo device, prefixALLOW_EXTRA_SIMULATOR=other-device; its messages never offer this. - Scripts:
test.sh(engine, then unit tests in two time zones) andui-test.sh(UI tests, each on a fresh install) run through sim-run.sh andclean-test-clones.shon exit;fresh-sim.sh(a true new install, onboarding from the start) runs on the demo device.
Devices and data
- No test data on a real device (Brad 2026-11-02). Demos, seeded records and UI-testing launches run in the simulator. Installing a build on a phone is fine; TestFlight is preferred.
- iPhone Mirroring can send a phantom mouse press before a tap. A tap-timing bug seen only through Mirroring is not proof of a user bug; a finger tap on the phone is (docs/gotchas.md, 2026-11-10).
- The simulator lies about some things: launch time (time it on a real iPhone, Release, cold) and background HealthKit delivery (the simulator never wakes the app). Some failures show only on a reinstall over an old install (the Keychain survives deleting the app).
Release
- Submit a ready build to TestFlight and App Review together, in one go; no phone test first (Brad 2026-11-09).
- An agent shell cannot upload (no App Store Connect token): it archives with
xcodebuild archive, then the session uploads from Xcode Organizer. Full steps: launch-checklist.md.
Gotchas
Silent ones to know before you start (full entries in docs/gotchas.md):
- 2026-11-02:
HKCorrelationTypeis not supported for background delivery; observe the quantity types. - 2026-10-23:
simctl spawn ... defaults writesurvives an uninstall;fresh-sim.shdeletes it too. - 2026-11-09: a stale App Attest key in the Keychain broke uploads and chat after a reinstall.
Search docs/gotchas.md for "iOS", "iPhone", "simulator" and "xcodebuild" for the rest.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

