feat: support iOS simulator transform gesture

[thymikee]

Summary

Add iOS simulator support for gesture transform.

This wires transform to a private XCTest synthesized multi-touch path for pan/scale, applies XCTest rotation for the rotation component, removes the old handler-level decomposition helper, and updates help/docs plus the command-planning coverage.

Closes #582

Validation

Ran pnpm format, pnpm build, pnpm build:xcuitest, pnpm exec vitest run src/utils/__tests__/args.test.ts, pnpm check:quick, git diff --check, and five transform math sanity variations.

Manual verification on iPhone 17 with Metro on 8082: gesture transform 201 421 80 -40 2 35 700 changed the app metrics from x 0, y 0, scale 1.00, rotate 0 with all change flags no to x 57, y -28, scale 1.56, rotate 23 with pan/pinch/rotate all yes. SkillGym was not run per request.

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://callstackincubator.github.io/agent-device/pr-preview/pr-586/
Built to branch gh-pages at 2026-05-26 07:53 UTC. Preview will be ready when the GitHub Pages deployment is complete.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2a1dda9fca

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[thymikee]

Reviewed the iOS simulator transform changes. Using XCSynthesizedEventRecord/XCPointerEventPath is a nice step up from the old pan+pinch+rotate decomposition, and the fallback to .unsupported(message) when the private API throws is the right call.

Main concerns:

1. The implementation is not one continuous two-finger gesture: pan+scale runs through the synthesized path (which ends with liftUpAtOffset:), and then rotation runs as a separate XCUIElement.rotate(...). That's at odds with the iOS smoke-suite contract you added ("Need one continuous two-finger gesture without lifting fingers"). Either:

   • Fold the rotation into the pointer path geometry (the Android helper already does this — interpolate θ(t) and place fingers at (centerX + cos(θ)*radius, centerY + sin(θ)*radius)), or
   • Update the contract + docs to acknowledge that rotation is a follow-up phase.

2. Hardcoded pinch radius (80 pt) isn't tunable and may not work well on iPad / for scale < 0.5. Either parameterize it or scale it off the app frame.

3. Robustness for private API drift: explicit nil checks on recordClass / pathClass and on each NSSelectorFromString would give clearer "Xcode version broke this" messages instead of an objc_msgSend crash.

4. Typedef width: RunnerMsgSendLongLong claims long long, but processID / interfaceOrientation are NSInteger. Works today on LP64 but the typedef should match the selector signature.

5. Codex bot's claim about the loop ending at t < 1.0 appears to be incorrect (the loop is <=). Suggest resolving that thread.

Nothing here is a hard block; the rotation-as-followup vs. atomic-gesture question is the one most worth aligning on before merge.

---

Generated by Claude Code