fix: speed up iOS text entry

[thymikee]

Summary

Fixes #1108 by removing unnecessary iOS text-entry waits while preserving the existing “type first letter, wait, then type the rest” behavior for reliable search fields.

Changes:

• add runner-side AGENT_DEVICE_RUNNER_TEXT_ENTRY_PHASE markers around focus, clear, type, warmup, verify, and total phases
• stop paying a second readiness wait after keyboard-visible focus has already been confirmed
• for concrete text targets, use XCUIElement.typeText so visible-keyboard refills stay scoped to the tapped field without a full focus timeout
• verify once on the normal success path and repair only after an observed mismatch

Measurements

Median wall-clock samples from iOS simulator runs:

Scenario	Before	After	Notes
Settings search fill, no settle	5890ms	2830ms	Removed duplicate post-focus wait and duplicate verify/repair.
Settings search fill, final settle comparison	3102ms no-settle	3973ms settle	Settle adds about 0.9s of stabilization.
Bluesky search replacement, no settle	5998ms	3594ms	Focus phase dropped from about 3.0s to about 0.57s; remaining large cost is clear-field work.
Bluesky search replacement, settle	6879ms	4523ms	Settle adds about 0.85s.
Bluesky empty search fill, no settle	n/a	2098ms	Runner phases are about 0.6s focus plus 1.18s type/verify.

Rejected experiments:

• private XCTest bulk typing selector: unavailable on Xcode 26.2, so the fast path never fired
• immediate hidden-keyboard target return: failed the Settings test bed, so it was not kept

Validation

• pnpm build:xcuitest
• git diff --check
• pnpm test-app:maestro:ios --close -- --udid C25DBB5B-9254-4293-A8D5-2785C78DE03A --state-dir /private/tmp/agent-device-test-app-maestro --artifacts-dir .tmp/issue-1108/maestro-ios --timeout 180000
  • The Checkout form Maestro flow passed in 22.8s.
  • The wrapper then exited non-zero only because its optional --close cleanup does not forward the custom state dir.

Follow-up note: repeated clean reruns of the stock Expo dev-client Maestro flow later hit test-app environment issues outside this text-entry path: launchApp clearState returned to the Expo dev launcher unless the runtime state was warm, and simctl get_app_container ... data eventually returned a missing simulator data-container path.

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	1.6 MB	1.6 MB	0 B
JS gzip	500.0 kB	500.0 kB	0 B
npm tarball	596.9 kB	597.8 kB	+898 B
npm unpacked	2.1 MB	2.1 MB	+5.1 kB

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	25.9 ms	25.9 ms	-0.0 ms
CLI --help	52.5 ms	52.3 ms	-0.2 ms

Top changed chunks: no changes in the largest emitted chunks.

[thymikee]

Reviewed the iOS text-entry speedup. The phase-marker logging, the keyboard-visible fast paths, and the switch to XCUIElement.typeText for concrete targets all look reasonable. One correctness concern on the verify/repair refactor — left it inline.

---

Generated by Claude Code

[thymikee]

Reviewed #1139 at head f862f5f against #1108. No actionable blockers found.

Checked:

• Scope is limited to the Swift runner text-entry path: RunnerTests+TextEntry.swift and focus/timing logging in RunnerTests+CommandExecution.swift.
• The duplicate post-focus readiness wait is removed only after a concrete target is tapped or after keyboard visibility already proves a bare first-responder path.
• Concrete-target typing now uses XCUIElement.typeText, while bare type still falls back to app.typeText when there is no resolved target.
• Replacement semantics still require a resolved target, keep clear-first behavior, retain first-character warmup, and verify with the existing stability window.
• Repair now happens only after an observed mismatch, then re-verifies the repaired value through the resolved element.
• No command-surface, daemon routing, interaction guarantee, or gesture/tap synthesis paths changed.
• CI is green across 21 checks after rerunning the transient Linux artifact-upload failure.

Residual risk: I did not rerun the simulator benchmark locally in this coordinator worktree; I am relying on the PR-provided Settings/Bluesky/test-app evidence plus green CI for maintainer judgment.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-07-07 07:50 UTC