Ios2android
Audit, implement, visually compare, and continuously update native Android parity from an existing iOS app, treating the current iOS product as the source of truth. Use when an AI coding agent such as Codex or Claude Code must port or mirror SwiftUI/UIKit screens into Kotlin/Jetpack Compose, detect and reproduce iOS commits that landed after the last verified Android sync, generate paired screenshot overlays, heatmaps, contact sheets, and reports, compare layouts and UI states, align feature/file structure, resolve cross-platform drift, or build and verify both apps in simulators/emulators.From its SKILL.md
npx -y skills add adraalabs/ios2android --skill ios2androidAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
12.5 KB, as published. Nobody here has run it
ios2android
Make Android express the same product as iOS. Translate intent, hierarchy, state, and interaction semantics; do not mechanically transliterate Swift into Kotlin or points into dp.
Hold the contract
- Treat iOS as read-only unless the user explicitly asks to change it.
- Preserve dirty worktrees and unrelated changes. Never reset, clean, or overwrite them.
- Read repository instructions before inspecting or editing either client.
- Use current iOS code, recent iOS commits, and a freshly built iOS runtime as evidence. Record conflicts between them.
- Keep backend and persisted-data contracts compatible. Do not repair visual parity by corrupting identity, dropping records, or changing API meaning.
- Match deliberate product behavior, not an obvious iOS bug. Surface ambiguous cases with evidence.
- Do not commit, push, deploy, or trigger cloud builds unless authorized.
- Never claim parity from compilation alone.
- Never advance the iOS sync checkpoint past an unimplemented, blocked, or unverified product change.
Run the parity workflow
1. Establish scope and safety
-
Locate the canonical iOS and Android roots, their repository boundaries, and applicable instruction files.
-
Inspect status, branch, remotes, submodules, and existing changes before editing.
-
Identify the iOS target/scheme, Android application module/variant, bundle ID, package name, minimum OS versions, installed SDKs, and available simulators. Record both the deployment baseline and newest practical iOS runtime.
-
Run the read-only inventory tool when the roots are discoverable:
python3 scripts/parity_inventory.py --ios /path/to/ios --android /path/to/android --output /tmp/parity-inventory.md -
Convert the request into observable acceptance criteria. Include every user-provided screenshot and reported failure.
2. Detect iOS work since the last Android sync
Do this at the start of every repeat parity run, even when the user names only one screen.
-
Determine whether “latest” means the current local iOS checkout or a remote tracking ref. If remote freshness is in scope, fetch the ref without merging, rebasing, or changing the iOS worktree.
-
Select and retain an immutable target iOS SHA for the run.
-
Check the tracked sync checkpoint:
python3 scripts/ios_delta.py check \ --ios /path/to/ios \ --android /path/to/android \ --head TARGET_SHA \ --output /tmp/ios-android-delta.md -
Process every pending commit oldest-first. Read complete diffs, including deletions, assets, localization, tests, schemas, dependencies, and project configuration—not only
.swiftfiles. -
Maintain a commit ledger mapping each iOS SHA and product intent to Android files, status, and verification evidence. Implement applicable changes; explain commits that are already equivalent or intentionally iOS-only.
If no checkpoint exists, establish the last defensible iOS revision already represented on Android. Never assume the current iOS HEAD is already synced. Read incremental-sync.md for bootstrap, rewritten-history, exit-code, and checkpoint rules.
3. Reconstruct the iOS source of truth
- Read recent iOS commit subjects and diffs. Start with UI, navigation, models, DTOs, themes, and removed-feature commits.
- Inspect the current Swift implementation of every affected screen and its shared components.
- Build and launch the current iOS source. Confirm the installed simulator build is the one just compiled.
- When standard iOS controls or materials can change with the SDK/runtime, also launch on the newest installed major iOS runtime. Treat runtime-adopted chrome such as tab bars, toolbars, sheets, and Liquid Glass as product evidence even when the Swift source did not change.
- Capture baseline states with stable data and note the source SHA, SDK, runtime, device, orientation, appearance, locale, authentication state, and content identity.
- Prefer evidence in this order when it conflicts:
- explicit user direction;
- current intentional iOS behavior built from source;
- current iOS implementation and tests;
- recent iOS commit intent;
- an older installed build or screenshot.
4. Build a parity matrix before broad edits
Map feature boundaries and screens, then audit each screen across these dimensions:
| Dimension | Inspect |
|---|---|
| Structure | feature folders, screen/view-model ownership, shared components |
| Navigation | tabs, pushes, sheets, dialogs, deep links, back behavior |
| Visual system | background, surface levels, borders, radii, spacing, type, icons, colors |
| Content | section order, labels, counts, metadata, posters, actions |
| Behavior | taps, long presses, menus, gestures, optimistic updates, permissions |
| State | loading, populated, empty, error, offline, stale cache, retry |
| Domain rules | watched/spoiler state, media identity, progress, removed features |
| Adaptation | phone/tablet, rotation, safe areas/insets, keyboard, RTL, long text |
| Accessibility | semantics, touch targets, contrast, dynamic text, reduced motion |
Read parity-method.md when planning a full-app audit or deciding whether a difference is intentional.
5. Translate SwiftUI to Compose semantically
- Align feature ownership before individual files. Mirror meaningful boundaries such as
Explore,Library,Feed,Person,Profile, andReviews; do not force one-to-one filenames where platform responsibilities differ. - Centralize shared tokens and primitives first: surfaces, hairlines, semantic accents, radii, buttons, cards, poster progress, spoiler shields, skeletons, and social actions.
- Preserve state ownership and lifecycle behavior. Reproduce what the user observes, including restoration, pagination, cancellation, and retry.
- Use native platform idioms when they preserve product semantics. Match hierarchy and emphasis rather than copying platform chrome pixel-for-pixel.
- For translucent system materials, distinguish foreground blur from backdrop sampling. Prefer a maintained native or Compose backdrop-effect implementation, verify its current stable release from primary documentation, keep a non-blur fallback, and test accessibility and rendering cost on the minimum Android API.
- Read swiftui-compose-map.md for translation patterns and common false equivalences.
6. Implement in vertical slices
Use this order unless dependencies require otherwise:
- design tokens and shared components;
- app shell and top-level navigation;
- primary tabs and their loading/empty/error states;
- detail routes, sheets, dialogs, and deep links;
- social posts, reviews, profiles, and editing flows;
- secondary settings/import/integration screens;
- responsive, RTL, accessibility, and offline passes.
For each slice:
- inspect the complete iOS state path;
- make the smallest coherent Android change;
- compile immediately;
- install the new APK, not a stale artifact;
- exercise the state on-device;
- compare against iOS and record the result;
- check crash logs before moving on.
Prefer shared fixes over repeated local overrides. Remove Android-only UI only after confirming the current iOS route and history intentionally removed it. Preserve underlying user data and compatibility behavior.
7. Verify both runtimes with equivalent evidence
-
Build both clients from their current source with isolated build output where practical.
-
Launch the exact built binaries on an iOS simulator and Android emulator.
-
Reproduce the same data identity and state on both platforms.
-
Capture paired screenshots with:
scripts/capture_pair.sh --bundle-id com.example.ios --package com.example.android --output /tmp/parity-shot -
Generate a visual report for each paired state:
python3 scripts/compare_screenshots.py pair \ --ios /tmp/parity-shot/ios.png \ --android /tmp/parity-shot/android.png \ --name "Feature · state" \ --output /tmp/parity-shot/report -
For a broad audit, build a JSON pair manifest and run
compare_screenshots.py matrixto create one interactive HTML index plus per-state reveals, overlays, heatmaps, blink views, masks, and JSON metrics. -
Expand every visible action that owns UI into a companion capture when practical: open avatar/edit badges, overflow and share menus, pickers, sheets, dialogs, context menus, and destructive confirmations. A resting screen cannot verify the options behind it.
-
Generate one labeled PNG containing every manifest pair for a fast whole-app scan:
python3 scripts/compare_screenshots.py contact-sheet \ --manifest /tmp/parity/visual-pairs.json \ --output /tmp/parity/all-screens.png -
Compare hierarchy first, then geometry, typography, color, imagery, and motion. For badges and controls that overlap artwork or avatars, also inspect clipping, z-order, edge overlap, icon contrast, and the full touch target. Treat the contact sheet and image metrics as triage signals, never as the parity verdict.
-
Test the failure paths that originally motivated the work, including deep links and content with missing or sentinel identities.
-
Inspect Android runtime crashes and iOS termination/fault logs.
-
Distinguish app defects from backend outages. Verify health independently, retain evidence, and resume the live pass after recovery.
Read visual-comparison.md before cropping, masking, setting thresholds, or interpreting image metrics. Read validation-playbook.md before the final simulator sweep.
8. Finish with evidence, not confidence
Before handoff:
- run whitespace/diff checks;
- run focused tests and the relevant full client builds;
- install and launch both final artifacts;
- rerun every reported regression;
- confirm no new runtime crash;
- review the complete parity matrix for untested cells;
- review the final visual matrix and retain paths to its HTML/JSON evidence;
- list any blocked or intentionally platform-specific differences;
- confirm unrelated changes were preserved;
- rerun the delta check against the selected target and then the latest intended iOS ref;
- if iOS advanced during the run, inspect and implement the new range before claiming Android is current;
- after all pending commits are implemented and verified, record the exact target with
ios_delta.py recordand include the checkpoint file with the Android changes.
Report:
- what now matches;
- which files/features changed;
- exact build and runtime checks performed;
- simulator/emulator identities;
- remaining differences with reasons;
- external blockers such as an unavailable API;
- whether anything was committed, pushed, or deployed;
- the previous checkpoint, target iOS SHA, new checkpoint, and per-commit dispositions.
Use report-template.md for a large audit. Do not use “everything matches” when any meaningful state remains unobserved.
Handle difficult cases
- Only screenshots are available: infer cautiously, inspect accessibility/UI trees, and mark behavior as unverified.
- The iOS app and source disagree: rebuild iOS, identify the installed version, and report the discrepancy before copying it.
- Backend data differs between platforms: compare requests, headers, account, locale, and payload contracts before changing UI.
- A post or media route fails: separate navigation identity from endpoint failure; never pass sentinel or legacy IDs as canonical media IDs.
- The API is down: verify independently, test cached/local states, and do not label HTTP failure as a crash.
- A platform control cannot look identical: preserve information architecture, prominence, and interaction semantics; document the native adaptation.
- The requested audit is very broad: keep one matrix and finish vertical slices. Avoid disconnected cosmetic edits that leave behavior drifting.