Advanced UI analysis and interaction through pixel-level reasoning to solve complex interaction problems, locate hard-to-find elements, verify visual states, and debug layout/rendering issues. Use as needed after UI Mapper inspection when advanced reasoning or non-standard views are required.
Scanned 9/6/2026
Install to Claude Code
npx -y skills add xAmirHamza77/ReverseOps-Skill --skill screenshots --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Screenshots?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/xamirhamza77-screenshots)More formats (shields.io, HTML) on the badges page.
---
name: screenshots
description: Advanced UI analysis and interaction through pixel-level reasoning to solve complex interaction problems, locate hard-to-find elements, verify visual states, and debug layout/rendering issues. Use as needed after UI Mapper inspection when advanced reasoning or non-standard views are required.
---
# Screenshots Skill (advanced mode)
> **Version**: 2.0 (Single-file mode) | **Backend**: Resolution-agnostic, uses pure pixel coordinates (pixels) | **Focus**: Advanced reasoning and problem solving
## CRITICAL: Microscope model (not primary analysis)
This skill is a **microscope, not a primary analysis tool**. It is Powerful for deep analysis, but Expensive (slow, high token usage). It Supports UI Mapper (the primary inspection tool).
**Key Principles:**
- **Less is More.** Use it SPARINGLY, NOT by default.
- **Flat Architecture.** Analyze in a single pass; NO recursive decomposition.
- **UI has no recursive structure.** Do not invent hierarchy where none exists.
- **Element Count Match.** MUST match the pixel/reference image as closely as possible.
- **Fail Fast.** If inspection fails after 1 retry, STOP and show the error.
- **No Features.** Do not document features/behaviors; ONLY report what you can SEE.
- **Debug-First.** Focus on malformed inputs, pollution, and state mismatches.
- **Advanced Analysis.** Here for complex analysis, not your FIRST analysis attempt.
- **Evidence Based.** Do not guess/hallucinate; base everything on observable evidence.
- **Pixel-Perfect ROI.** Validate element sizes through actual screenshot regions, NOT via uamappings alone (accuracy gate).
---
## CRITICAL: Data freshness (non-negotiable)
- Inventory Data MUST be < 30 seconds old. If `since_capture` > 30s, RE-CAPTURE.
- Screenshots MUST be from the CURRENT session (after the last capture/screenshot command).
- NEVER reuse uamappings across contextual changes (app switch, navigate, resize, session change).
- Changing target app invalidates prior inventory data (target_mismatch).
- After device capture, ALWAYS wait 300–500ms before screenshot ROI processing.
- NEVER assume the UI is static; verify with a fresh capture if unsure.
- If the map is Stale, RE-CAPTURE (fresh screenshot + `uiauto inventory + map`) before proceeding.
---
## CRITICAL: Screenshot performance (keep files small)
- NEVER use Full Full-Screen Screenshots — wasteful and unnecessary.
- ALWAYS use Region Capture (ROI) — capture only the region of interest.
- ROI Validation: Capture → Compare. MUST be high-confidence; retry if not.
- Save all screenshots to: `screenshots/[app-name]-[feature]/` (NEVER the bundle root).
- Filename Pattern: `[NNN]-[slug]-[focus].png` (e.g., `001-search-entry-input.png`).
- Sequential numbering is scoped per screenshot folder (each new folder starts at 001).
- Slug: Matches the folder slug (e.g., `search-entry`, `app-tab-bar`).
- Focus: Describes the ROI of the image (e.g., `toolbar`, `dialog`, `list`).
- If saving fails, halt with an ERROR (screenshot unsalvageable).
---
## When to use which tool
**Use UI Mapper** for standard inspection (fast and efficient).
**Use Screenshots** for these hard problems:
| Use screenshots when: | Example |
|------------------------------|---------------------------------------------------|
| Elements are hard to find | Element not identified by uamappings/automation |
| Complex interactions needed | Multi-step, context menus, drag-and-drop |
| Visual state validation | Element state (enabled/disabled, checked, colors) |
| Layout/rendering debugging | **O**verlaps, alignment, cut-off text, resizing |
| Non-standard views | Canvas, OpenGL, custom-drawn, games |
| Pixel-level precision required | Bounds off by even one pixel |
| Understanding visual structure | Component layout and relationships |
---
## System instructions: what you are allowed to do
You have Permission to Pause/Wait/Inspect during interactions for review of the updated uamappings and flatmap to keep the interaction context in focus.
**Form factors**: Desktop, Mobile, Smart TV share the same XML structure (device-agnostic).
**Supported actions and gestures:**
- **Basic gestures**
| Action | Android | iOS | Desktop |
| ------------------------------ | ------------------ | ----------------- | ------------------ |
| **Tap/Click** | `tap {x} {y}` | `tap {x} {y}` | `click {x} {y}` |
| **Long Press/Right-click** | `long_press {x} {y} [ms]` | `long_press {x} {y}` | `rightclick {x} {y}` |
| **Double Tap/Double-click** | `2 × tap` | `2 × tap` | `doubleclick {x} {y}` |
| **Type** | `type "text" {x} {y}` | `type "text" {x} {y}` | `type "text" {x} {y}` |
- **Scrolling and navigation**
| Action | Android | iOS | Desktop |
| --------------------------- | -------------------- | ---------------------- | -------------------- |
| **Scroll** | `swipe {x1} {y1} {x2} {y2} [ms]` | `scroll {x} {y} [distance]` | `scroll {x} {y} [up/down] [clicks]` |
| **Horizontal swipe** | `swipe {x1} {y1} {x2} {y2}` | `swipe {x1} {y1} {x2} {y2}` | `drag {x1} {y1} {x2} {y2}` |
- **Mouse-specific (Desktop only)**
| Action | Description |
| ------------------------------- | --------------------------------------- |
| `move {x} {y}` | Move the cursor to `{x} {y}` without clicking |
| `drag {x} {y} {to_x} {to_y}` | Click-and-drag from `{x} {y}` to `{to_x} {to_y}` |
| `rightclick {x} {y}` | Context menus (macOS `rightclick`, Windows `rclick`) |
| `doubleclick {x} {y}` | Double left-click (Desktop) |
- **Other platform-specific actions**
| Action | Description |
| ---------------------------------------------- | ---------------------------------- |
| `keyevent <KEY>` | Android: back, home, recent apps, etc. |
| `press_key "KEY"` | Windows: press a specific key |
| `launch <bundle-id>` | Launch an app |
| `hide` | Hide (macOS only) |
- **Gestures (Advanced)** — Multi-touch on Mobile
| Gesture | Platform | Description |
| -------------------------------- | -------- | ----------------------------------- |
| `pinch {x} {y} [scale] [steps]` | Mobile | Zoom in/out at `{x} {y}` |
| `rotate {x} {y} [rotation] [steps]` | Mobile | Rotate at `{x} {y}` (degrees) |
| `drag_and_drop {x1} {y1} {x2} {y2} [ms]` | Mobile | Press, hold, and move coordinates |
---
## Core workflow (5 steps)
Use cases (troubleshooting, exploration, validation) share the same workflow:
### Step 1: check if this skill is actually needed (decision gate)
Before diving into pixels, ask one question:
> **"Did UI Mapper give me enough info to proceed?"**
- **YES** → Stop here. Use UI Mapper (it is 10x faster).
- **NO** → Continue, and read `references/microscope-later.md` to understand when to step back.
### Step 2: identify exact screen/bounds (discovery)
**NEVER guess positions. Derive coordinates from uamappings data.**
1. Capture at the system level (NOT uamappings): `uiauto capture --output-dir ./shots/`
2. Extract the screen/bounds of the target element:
- Look for `<window>` or bounds data in the initial XML.
- Note the full screen position from the FILE header (Screenshot: `path`).
3. You may need the actual Pixel coordinates later for ROI/pixel validation.
### Step 3: record success and failure states (evidence)
Automated verification returns JSON (`success: true/false`). You MUST map this to the VISUAL state.
| CANONICAL | Visual (Color) | Visual (Form) | Visual (Opacity) | Note |
| --------------- | ------------------- | -------------- | ---------------- | ---- |
| `disabled` | Gray (#9E9E9E) | Faded icon | ~40—60% | Cannot click |
| `read-only` | Light gray | Static text | 100% | May show a caret |
| `processing` | Blue/Orange | Spinner/Progress | Animated | Busy |
| `failed` | Red/Yellow | Icon | Highlighted | Action required |
| `pending` | Neutral | Dashed border | ~80% | Waiting for input |
**ALWAYS capture evidence of BAD, ERROR, MALFORMED, or DEVIATING states FIRST before continuing.**
---
### Step 4: analyze and interact (pixel-level interaction)
- **Flat description**: What you see (visuals, bounds, hierarchy).
- **High-confidence ROI for interaction**:
No blind coordinate guessing. Determine target coordinates 100% from uamappings (min 5 targets).
- **Validate bounds** (zero margin):
Check that click bounds are correct. ALWAYS click the center (50% point).
- **Zoom via cropping**: Use ROI screenshots of relevant regions instead of full-site work.
- **Bug/challenge**: Address what you SEE (not what you Think). Fix before continuing.
- **Cross-validation**: Compare XML semantic data (`<clickable>`, `<edit-text>`) against pixel evidence.
**Click validation:**
- ALWAYS Click the Center of the target element.
- Check if the Click landed (visual change or uiauto event).
- If Click Failed → Adjust accuracy (bounds validation), NEVER click randomly elsewhere.
- Max 2 attempts; if it still fails → STOP and assume the environment is Broken.
---
### Step 5: cleanup, lints, and finish
- Keep your Screenshots/Summary in the current folder (`./screenshots/[app-name]-[feature]/`).
- Fill in the Reporting Template (see File Pointers below).
- Do not guess; document only What you have Seen.
---
## Handling failure (self-correction)
### Common failure modes
| Symptom | Cause | Fix |
| ------------------------- | ------------------- | --------------------------- |
| Capture returns empty XML | System/UI permission | Check accessibility permissions |
| Element not in tree | Rendering/Shadow DOM | Use screenshots ROI directly |
| Bounds look wrong | Scaling factor (Retina/Display) | Check `uamappings` for the ratio |
| Interaction timeout | Animation running | Increase wait to 500ms before clicking |
### Debugging flow (Fail Fast)
1. WAS the step successful (JSON return); if False, WHY.
2. WHAT evidence can you gather: fresh capture, new uamappings, ROI screenshots.
3. WHOSE fault is it — application or environment?
4. WHERE in the tree is the failure?
5. WHEN did it fail (timing/context)?
6. HOW do we fix it (re-capture, wait, ROI)?
---
## Analysis checklist (self-validation)
- [ ] Did you read `references/microscope-later.md`?
- [ ] Did you enforce Data Freshness (<30s)?
- [ ] Did you verify the uamappings (Internal Review)?
- [ ] Did you record failure states (Evidence collection)?
- [ ] Did you use flat analysis (NO recursive decomposition)?
- [ ] Did you click at 100% confidence (ALWAYS center)?
- [ ] Did you minimize file artifacts (ROI instead of full screen)?
- [ ] Did you stay in character (Microscope supporting UI Mapper)?
---
## Visual element descriptions (flat analysis)
```text
[Component Name] - [VISUAL STATE] - [INTERACTION]
[Specific visual details - what you literally see]
Text content: "[Exact visible text]"
Colors (if relevant):
- Background: #RRGGBB or name (gray, blue)
- Text: color/weight
- Border/shadow: if present
Dimensions/Spacing:
- Width x Height: element bounds
- Spacing: gaps to neighbors, alignment
Visual indicators:
- Icons: name/description
- Shadows: elevated/floating appearance
- Borders: rounded corners, outline thickness
Interaction states (if visible):
- Default: normal appearance
- Hover: color/underline changes (desktop only)
- Active/Pressed: highlighted appearance
- Disabled: grayed out, cursor not-allowed
```
---
## Reporting template
- **Investigation goal:** [What was I looking for?]
- **Screenshot ID:** [Filename/Sequence #]
- **Findings:** [Structure, Elements, Micro-interactions]
- **Data quality:** [Confidence percentage]
- **Conclusion:** [Is the Goal reached?]
## Example analysis output
```text
[Example Analysis] — [short title]
WHAT I ANALYZED:
- Target: [component name and type]
- Platform: [iOS/Android/Desktop]
- Method: [screenshot analysis, bounds validation, etc.]
STRUCTURE:
- Layout: [flat visual layout]
- Hierarchy: [parent-child relationships]
INTERACTIONS:
- Touch targets: [okay/constraints]
- Gestures: [tap, scroll, etc.]
FINDINGS & ISSUES:
- [Component A]: [finding]
- [Component B]: [finding]
TARGETS FOR INTERACTION (with coordinates):
- [Element 1]: ({x}, {y}) width x height - [action]
- [Element 2]: ({x}, {y}) width x height - [action]
CONCLUSION:
[Key finding or main insight from this analysis]
```
---
---
## Platform-specific reference guide (CRITICAL: read only per issue)
This file does NOT load platform-specific reference files by default (to minimize context). You MUST consult the correct reference for your task to avoid repeated attempts.
### When to read a reference
| Your task | Reference |
|--------------------------|---------------------------------------------|
| Element not found | `troubleshooting.md` |
| Interaction failed | `troubleshooting.md` |
| Performance/timeout | `troubleshooting.md` |
| Looking for coordinates | `bounds.md` |
| Drag-and-drop/swipe | `element-types.md` |
| Platform-specific behavior | `interactions/{platform}.md` |
| Efficiency/cost questions | `efficiency.md` |
### Critical issue patterns
**Element not found?**
→ Read: `references/troubleshooting.md` (element type patterns) + `references/bounds.md` (validation)
**Actions failing?**
→ Read: `references/troubleshooting.md` (type mapping matrix) + `interactions/{platform}.md`
**Multiple apps open?**
→ Verify: `uiauto focus` first, target the CORRECT app.
**Fullscreen game/canvas?**
→ Read: `references/element-types.md` (canvas regions, OpenGL) + `references/troubleshooting.md`
---
## Quick start by issue
| Issue | Action |
| ----------------------- | --------------------------------------------------------------------- |
| **Element not found** | Read `troubleshooting.md` (element type patterns) |
| **Click/typing failed** | Read `troubleshooting.md` (type mapping) + `interactions/{platform}.md` |
| **Performance/timeout** | Read `troubleshooting.md` (validation) + `efficiency.md` |
| **Multiple apps open** | Verify: `uiauto focus --target [app-name]` first |
| **Fullscreen game?** | Read `element-types.md` (canvas regions) + `troubleshooting.md` |
| **Non-standard UI?** | Read `element-types.md` (custom views, video, WebView) + `troubleshooting.md` |
---
## Element classification guide
### Navigation (structural)
**Role**: Present for orientation or accessing content on other screens (less important).
| Element | Type | Signals |
|-----------------|------------------|------------------------------------------------|
| Button | **navigation** | Top-left, back-arrow icon |
| Tab | **navigation** | Fixed at bottom, tab role, links to other screens |
| Link | **navigation** | Colored text, "`→`" icon, changes screen |
| Sidebar item | **navigation** | Left on desktop, navigation role |
### Controls (primary action)
**Role**: Things the user **taps** to get something done.
| Element | Type | Signals |
|-------------|-------------|--------------------------|
| Button | **control** | high z-order, prominent |
| Input field | **control** | edit text, placeholder |
| Slider | **control** | thumb/gutter, selectable |
| Toggle | **control** | on/off state |
| FAB | **control** | Floating Action Button |
### Content (information)
**Role**: Content the user consumes (read, watch, listen).
| Element | Type | Signals |
|---------------------------------|-----------|-------------------------------|
| Textview | **content** | role=text, not clickable |
| Image | **content** | Visual, not a button |
| Header | **content** | Semantic header |
| List item | **content** | In list, between other items |
| Label/description | **content** | role=text |
### Decorations (ambient)
**Role**: Visual only, no interaction.
| Element | Type | Signals |
|----------------|---------------|-----------------------------|
| Spacer/divider | **decorations** | Empty, role=separator |
| Section bg | **decorations** | Background color only |
| Decorator icon | **decorations** | Purely aesthetic |
| Padding | **decorations** | Absolutely no padding |
### Functional elements (CRITICAL: below containers)
**Role**: Container with **meaning** that affects interaction.
| Element | Type | Signals |
|--------------------|--------------|-------------------------|
| Modal overlay | **functional** | dimming behind, z-index |
| Search bar | **functional** | Search input in nav |
| Progress tracker | **functional** | Visual progress over steps |
| Active tab | **functional** | Visual difference vs other tabs |
| Toolbar | **functional** | Fixed top, tools for current |
---
## Anchors: only for interactions (not all elements)
**Anchors** are reference points for actions/interactions (tap, click, etc.).
### When you need an anchor
✅ **Interactive** (needs an anchor): Button, SearchField, Tab, InputField
❌ **Visual** (no anchor): Background, Image, Label-only, Spacer
### How to make an anchor (device-agnostic)
```text
SELECT: [element role or type]
WHERE: [property = value, scope = parent]
```
**Examples:**
- iOS Tab: `SELECT: AXButton WHERE: label = "Search" IN scope = TabBar`
- Android Button: `SELECT: button WHERE: text LIKE '%Save%'`
- Desktop Search: `SELECT: AXTextField WHERE: placeholder CONTAINS 'Search'`
---
## Role mapping across platforms
| Role | iOS (SwiftUI/UIKit) | Android (Compose) | Desktop (Win/Mac) |
| ------------------- | ------------------------------- | ----------------------------- | ------------------------ |
| **button** | AXButton | button | Button / AXButton |
| **text field** | AXTextField / AXSecureTextField | edit-text | TextField / Edit |
| **static text** | AXStaticText | text-view | Text / AXStaticText |
| **image** | AXImage | image-view | Image / AXImage |
| **link** | AXLink | text-view (clickable) | Hyperlink |
| **checkbox** | AXCheckBox | check-box | CheckBox |
| **radio button** | AXRadioButton | radio-button | Radio |
| **slider** | AXSlider | seek-bar | Slider |
| **switch/toggle** | AXSwitch / AXToggle | switch | Toggle / AXSwitch |
| **progress bar** | AXProgressIndicator | progress-bar | ProgressBar |
| **tab** | AXTabBar / AXTabGroup | tab | Tab / TabItem |
| **list** | AXTable | recycler-view | List / ListBox |
| **menu** | AXMenu | menu | Menu / ContextMenu |
| **heading** | AXHeading | text-view (large) | Text (styled) |
---
## Element interaction mapping
| Android node type | Interaction type | Available actions |
| ------------------- | ---------------- | --------------------------- |
| `text-view` | content (read) | N/A (read-only) |
| `edit-text` | control | tap (focus), type text, long-press |
| `button` | control | tap, long-press (if enabled) |
| `image-button` | control | tap, long-press |
| `check-box` | control | tap (toggle) |
| `switch` | control | tap (toggle) |
| `seek-bar` | control | tap, swipe (adjust) |
---
## Bounds precision (CRITICAL for accurate interaction)
**Golden rule:** Bounds are **admicrosikropisch** (limits, not suggestions).
- Element correctly defined: `<x>D` bounds, not your interpretation.
- Deviation of even one pixel can make clicks worthless (Android tap = 64px target minimum).
**Why bounds matter:**
- **Interaction accuracy**: Tap at (100, 200) succeeds if the element is large and this is safe; for a small button (24 x 24px), a one-pixel error wastes the tap.
- **State detection**: `enabled="false"` elements are NOT clickable even if they look clickable. Bounds + state = safe interaction.
- **Zooming/magnification**: Magnifier mode (Android, accessibility service, computational zoom) is NOT a solution; the user's view is not your screenshot data.
**Common format:**
```text
<x> y <width> height
[123] 456 [789 x] 90
↑ X-coordinate (px from left)
↑ Y-coordinate (px from top)
↑ Width (pixels)
↑ Height (pixels)
```
---
## Element count match requirement
**CRITICAL:** Your element count MUST match the screenshot as closely as possible.
- Count every **meaningful interactive** element (not controls).
- If you see 8 buttons, list 8 buttons (not 7, not 9).
- If some are disabled or off-screen, note it ("8 buttons, 1 disabled").
- Empty containers/spacers: add them as decorations but not counted.
- Use Report: count: (e.g., "Found 5 buttons matching description")
---
## Failure modes & debugging
**"Element not found" in uamappings?**
1. Check **element count match** — are you looking for too many/few?
2. Is it a **content** element? Non-interactive items (spacers) may be filtered.
3. Is it a **custom view** (OpenGL, Canvas)? Use screenshots and manual bounding boxes.
4. Is the UI **below the fold**? Scroll to reveal it.
**"Wrong platform behavior" (desktop feels like mobile)?**
1. Check platform in the uamappings header.
2. Forms behave differently (desktop = mouse/keyboard, mobile = touch).
**"Interaction timeout/animation"?**
1. Wait 500—1000ms before scrolling again.
2. Use explicit `uiauto wait 500` commands.
3. Or capture screenshot + uamappings immediately.
**"Cannot find the search bar"?**
1. Search bars are often **functional elements** (not just controls).
2. They may be in navigation/toolbars (not in the main container).
3. Check the uamappings: they may appear as `edit-text` + `search` hint.
```text
[Error Handling]
- Element not found → Retry up to 1 time → STOP and report [ERROR]
- Bounds mismatch → Re-capture; use ROI screenshots to confirm actual bounds
- Multiple targets → Try to focus on one app, or local context if needed
```
---
## Core anti-hallucination guard: "What do you see, not what you expect"
**Rule:** Describing what you CANNOT SEE is a [HALLUCINATION].
- Do not add features, tooltips, or menus that are not visible.
- Describe tint colors, contrast, weights — any pixel you see.
- Use VISUAL evidence colors: `"Dark Blue (#0B3D91)"`, not just "blue".
---
## Role in the skill system
You are a **lens tool**, not the analysis tool.
**Relationship to other skills:**
- **ui-mapper**: Standard inspector (fast). Use it FIRST for overview.
- **interaction**: Uses your anchor results to perform actions.
- **agent-orchestrator**: Invokes you (the advanced) when needed.
**Here is where you fit:**
```text
Task → UI Mapper (overview) → Screenshots (deep dive if needed) → Interaction (act)
↓ ↓
(fast, 47% token) (slow, zoom)
```
---
## Platform-specific considerations (selective)
**iOS:** Prefers UIKit patterns; check SF Symbols (system icons); text placeholder in AXTextField.
**Android:** Wide variation by manufacturer; check Intent actions (`android.intent.action.VIEW`), gestures like swipe-to-refresh.
**Desktop:** Menu bar, window chrome, draggable regions, keyboard shortcuts (Cmd/, Ctrl+), mouse hover states, scroll wheel.
---
## Efficient operations (last guidance)
| Operation | When | Token cost |
| ----------------------- | ----------------------------- | ------ |
| **Capture screen** | When map is stale (>30s) | ~400 |
| **Region capture (ROI)** | Most analysis (recommended) | ~800 |
| **Full screenshot** | Only for small screens | ~2000+ |
| **Bounds validation** | Verification needed | ~200 |
| **Element search** | One element or tree | ~300 |
**Token efficiency goals:**
- FAST: < 800 tokens per analysis cycle
- Acceptable: < 1500 tokens
- Do NOT exceed this max: > 2000 tokens (warn against repeated heavy captures)
---
## Critical confidence signals
| Signal | Confidence | Action |
| ----------------------------- | ----------- | ------------------------------------- |
| **Visible text** matches hint | 95% | Proceed: element is the intended one |
| **Role matches expected** | 90% | Consider bounds size |
| **Interactive** | 85% | Verify bounds match the action type |
| **Attributes correct** (disabled?) | 100% | Stop: element NOT usable for the action |
---
## Final rule
**If you cannot see it, you cannot click it.**
**If you cannot measure it (pixels), do not attach precision to it.**
**If you cannot verify it, do not assume it.**
---
This skill is the final authority when you need **pixel-perfect clarity**.
---
## File pointers
| Reference file | When to read |
| -------------------------------------| ---------------------------------------------------- |
| `references/microscope-later.md` | [MUST] First-time usage — read before first analysis |
| `references/troubleshooting.md` | Elements not found / interaction failed |
| `references/bounds.md` | Coordinates / bounds questions |
| `references/element-types.md` | Element classification / role mapping questions |
| `references/efficiency.md` | Token cost/efficiency validation |
| `interactions/{platform}.md` | Platform-specific behavior (ios/android/windows/macos) |
| `references/flatmap-structure.md` | Flat analysis UI structure / composite patterns |
| `references/debugging-flow.md` | Failure modes / self-correction |
**Rule:** Read `microscope-later.md` first, check platform interactions as needed.
---
## Version & accuracy
**Version**: 2.0 (Single-file mode)
**Backend**: uamappings + ROI pixel validation (accuracy > 95%)
**Scale**: Resolution-agnostic via flat structural paths
**Last updated**: [Current date]
**Accuracy gate**: All bounds come from actual pixels + uamappings verification, NOT from uamappings alone.
[IMPACT:ACTIVE:ImpactAnalysis/active]
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!