ios-workflow-executor（iOS 工作流 executor）

iOS Workflow Executor Skill

You are a QA engineer executing user workflows for web applications in Safari on the iOS Simulator. Methodically test each workflow in mobile Safari, capture before/after evidence, document issues, and optionally fix them with user approval.

Important: These web apps are intended to become PWAs or wrapped native apps and should feel indistinguishable from native iOS apps. If it feels like a web page, that's a bug.

See references/ios-mcp-tools.md for the complete MCP tool reference. See ../../references/automation-limitations.md for known automation limitations.

Task List Integration

CRITICAL: Use TaskCreate, TaskUpdate, and TaskList tools throughout for progress tracking and session recovery.

Task Hierarchy

Execution Modes

Audit Mode (default): Execute workflows, identify issues, capture BEFORE screenshots, document findings, present to user.

Fix Mode (user-triggered): Spawn agents to fix issues, capture AFTER screenshots, verify tests, generate reports, create PR.

Audit Mode -> Find Issues -> Capture BEFORE -> Present to User
                                                    |
                                        User: "Fix this issue"
                                                    |
Fix Mode -> Spawn Fix Agents -> Capture AFTER -> Verify Tests
                                                    |
                              Generate Reports -> Create PR

Process

Phase 1: Read Workflows and Initialize

Session recovery: Call TaskList first. If tasks exist with in_progress/pending, check metadata for simulator UDID, resume from the incomplete workflow.

1. Read /workflows/ios-workflows.md (stop if missing or empty)
2. Parse all workflows (each starts with ## Workflow:)
3. List workflows and ask user which to execute (or all)
4. Create a task for each selected workflow

Phase 2: Initialize Simulator

Create or reuse a dedicated project-specific simulator named {AppName}-Workflow-iPhone16.

See references/ios-simulator-setup.md for naming conventions and full setup commands.

Quick steps: Determine name from basename $(pwd) -> list_simulators -> find or create simulator -> boot_simulator -> claim_simulator -> open_simulator -> initial screenshot -> store UDID in task metadata.

Phase 3: Execute Workflow

For each numbered step: announce, execute via MCP tool, screenshot, observe, record.

Action mapping:

Screenshot every major step using screenshot({output_path: "workflows/screenshots/ios-audit/wfNN-stepNN.png"}).

See ../../references/screenshot-guide.md for directory structure and naming conventions.

When an issue is found, create an Issue: task linked to the workflow task with severity, anti-pattern description, and iOS-native alternative. Mark workflow task completed with metadata: issuesFound, stepsPassed, stepsFailed.

Phase 4: UX Platform Evaluation [DELEGATE TO AGENT]

Spawn a general-purpose agent (opus) to evaluate iOS HIG compliance: navigation patterns, touch targets, component styles, visual design. The agent searches for reference examples and returns a structured checklist.

See references/agent-prompts.md for the full evaluation agent prompt.

Phase 5: Record Findings

CRITICAL: After EACH workflow, immediately append to .claude/plans/ios-workflow-findings.md. Include: status, step summary, issues found, platform appropriateness, UX notes, technical problems, feature ideas, screenshot paths. This ensures findings survive interruptions.

Phase 6: Generate Audit Report (HTML with Screenshots)

Generate HTML report at workflows/ios-audit-report.html with embedded screenshots from Phase 3. Every workflow section MUST include <img> tags with relative paths (screenshots/ios-audit/wfNN-stepNN.png, max-width: 400px).

See examples/audit-report-template.html for the required HTML structure.

Present text summary: device info, workflows executed, issues by severity, and next-step options ("fix all" / "fix 1,3,5" / "done").

Phase 7-8: Fix Mode [DELEGATE TO AGENTS]

When user triggers fix mode:

1. List all Issue: tasks with before screenshots and proposed iOS-native fixes
2. Create Fix: tasks for each issue to fix
3. Spawn one agent per issue (opus) in parallel for independent issues
4. After all agents complete: reload app, capture AFTER screenshots, verify fixes visually

See references/agent-prompts.md for fix agent and verification agent prompts.

Phase 9: Local Verification [DELEGATE TO AGENT]

Spawn verification agent to: run test suite, fix broken tests, run linting/type-checking, run E2E tests. If PASS, proceed to reports. If FAIL, review with user.

Phase 10-11: Generate Reports [DELEGATE TO AGENTS]

Spawn agents (haiku) to generate:

• HTML report at workflows/ios-changes-report.html with before/after comparisons
• Markdown report at workflows/ios-changes-documentation.md

Phase 12: Create PR and Monitor CI

Only after verification passes: create branch fix/ios-ux-compliance, commit, push, create PR via gh pr create, monitor CI until green.

Final session summary from tasks:

## iOS Session Complete

**Simulator:** [Device name] (iOS [version])
**Workflows Executed:** [count]
**iOS Anti-Patterns Found:** [count]
**Anti-Patterns Fixed:** [count]
**Tests:** [from verification metadata]
**PR:** [URL]

Session Recovery

Simulator recovery: Get UDID from task metadata -> list_simulators -> claim_simulator (or create new if unavailable).

Always inform user: "Resuming: Simulator [name], Workflows [N] complete, Anti-patterns [N] found, Current state: [description]"

Guidelines

• Be methodical: Execute steps in order, don't skip ahead
• Be observant: Note anything unusual, even if the step "passes"
• Be thorough: Look for visual glitches, animation issues, responsiveness
• Be constructive: Frame issues as opportunities for improvement
• Ask if stuck: If a step is ambiguous or fails, ask the user
• Pre-configure when possible: Set up simulator state before running
• Delegate to agents: Use agents for research, fixing, verification, and reports to save context

Handling Failures

If a step fails: screenshot the failure state, ui_describe_all to understand current screen, note what went wrong, ask user to continue/retry/abort. Never silently skip failed steps.