{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"github-x-plug--mobileagent","slug":"x-plug--mobileagent","name":"MobileAgent","type":"agent","url":"https://github.com/X-PLUG/MobileAgent","page_url":"https://unfragile.ai/x-plug--mobileagent","categories":["ai-agents"],"tags":["agent","android","app","automation","copilot","gui","mllm","mobile","mobile-agents","multimodal","multimodal-agent","multimodal-large-language-models"],"pricing":{"model":"open_source","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"github-x-plug--mobileagent__cap_0","uri":"capability://image.visual.multimodal.gui.perception.and.element.grounding","name":"multimodal gui perception and element grounding","description":"Uses GUI-Owl vision-language models (1.5, 7B, 32B variants) built on Qwen3-VL to perform native visual understanding of mobile/desktop UI elements and generate precise bounding box coordinates for detected components. The model unifies perception, grounding, and reasoning in a single forward pass, enabling pixel-accurate element localization without separate object detection pipelines or post-processing heuristics.","intents":["I need to identify and locate specific UI elements in a mobile app screenshot to automate interactions","I want to understand the visual layout and semantic meaning of GUI components without manual annotation","I need bounding box coordinates for elements to drive automated clicks and interactions"],"best_for":["mobile app automation engineers building cross-platform test suites","GUI automation framework developers needing vision-based element detection","teams automating Android/HarmonyOS/desktop workflows without accessibility APIs"],"limitations":["Model inference latency varies by variant (1.5 faster than 7B/32B); larger models provide better reasoning but slower grounding","Grounding accuracy degrades on heavily obfuscated or custom-rendered UI elements not well-represented in training data","Requires GPU for reasonable inference speed; CPU-only inference adds 2-5x latency overhead"],"requires":["GUI-Owl model weights (1.5, 7B, or 32B variant) from Alibaba/Tongyi Lab","Python 3.8+","CUDA 11.8+ for GPU acceleration (optional but recommended)","Mobile device or desktop with ADB/Playwright/PyAutoGUI connectivity for screenshot capture"],"input_types":["PNG/JPEG screenshots from mobile devices or desktop","Raw pixel data from frame buffers","Video frames for continuous UI monitoring"],"output_types":["JSON with element bounding boxes [x, y, width, height]","Structured UI element descriptions with semantic labels","Confidence scores per detected element"],"categories":["image-visual","perception-grounding"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_1","uri":"capability://planning.reasoning.task.planning.and.multi.step.action.decomposition","name":"task planning and multi-step action decomposition","description":"Implements hierarchical task planning using GUI-Owl reasoning capabilities to decompose high-level user intents into sequences of atomic GUI actions (tap, swipe, type, scroll). The framework uses explicit thinking chains (Thinking variants of GUI-Owl) to generate step-by-step action plans with intermediate state validation, enabling recovery from partial failures and dynamic replanning when UI state diverges from expectations.","intents":["I need to break down a complex user workflow (e.g., 'send a message to John') into individual app interactions","I want the agent to reason about preconditions and dependencies between actions before executing them","I need the agent to adapt its plan when the UI doesn't match expected state after an action"],"best_for":["automation engineers building multi-step mobile app workflows","QA teams automating complex user journeys across multiple screens","developers creating self-healing test automation that adapts to UI changes"],"limitations":["Planning latency increases with task complexity; 5-10 step tasks typically require 2-5 seconds of model inference","Reasoning quality depends on model variant; GUI-Owl-1.5 provides faster planning but GUI-Owl-32B offers more robust decomposition for ambiguous intents","No built-in persistent memory across sessions; each task planning starts fresh without learning from previous executions"],"requires":["GUI-Owl model with Thinking capability enabled","Python 3.8+","Mobile-Agent-v3.5 or v3 framework","Connected Android device or desktop environment for action execution"],"input_types":["Natural language task description (e.g., 'open Settings and enable WiFi')","Current UI screenshot for context","Optional: previous action history for replanning"],"output_types":["Structured action sequence with [action_type, target_element, parameters]","Reasoning trace explaining the decomposition logic","Confidence scores for each planned action"],"categories":["planning-reasoning","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_10","uri":"capability://data.processing.analysis.evaluation.and.benchmarking.on.standardized.mobile.automation.tasks","name":"evaluation and benchmarking on standardized mobile automation tasks","description":"Provides comprehensive evaluation framework with standardized benchmarks (GroundingBench, GUIKnowledgeBench) to measure agent performance on mobile automation tasks. Metrics include action success rate, task completion rate, action efficiency (steps to completion), and grounding accuracy. Enables reproducible comparison across agent versions and model variants.","intents":["I want to measure and compare agent performance across different model variants and framework versions","I need standardized benchmarks to evaluate automation quality and identify improvement areas","I want to track performance improvements over time as the system evolves"],"best_for":["researchers evaluating mobile automation systems","teams tracking performance metrics across development cycles","organizations comparing different automation approaches"],"limitations":["Benchmarks may not reflect real-world app diversity; performance on benchmarks doesn't guarantee performance on unseen apps","Evaluation requires running full automation workflows which is time-consuming; benchmark suites may take hours to complete","Metrics like 'task completion' require manual definition of success criteria; different definitions lead to incomparable results","Benchmark apps may change over time (app updates, UI redesigns) making historical comparisons invalid"],"requires":["Mobile-Agent-v3.5 framework with evaluation utilities","Python 3.8+","Connected Android device or emulator","Benchmark app APKs (provided in repository)"],"input_types":["Task descriptions and success criteria","Benchmark app APKs","Agent configuration (model variant, parameters)"],"output_types":["Structured evaluation results (JSON/CSV)","Performance metrics (success rate, efficiency, grounding accuracy)","Detailed execution traces for failed tasks","Comparison reports across versions"],"categories":["data-processing-analysis","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_11","uri":"capability://text.generation.language.natural.language.task.specification.and.intent.understanding","name":"natural language task specification and intent understanding","description":"Accepts high-level natural language task descriptions (e.g., 'send a message to John saying hello') and uses GUI-Owl reasoning to understand user intent, extract key entities and constraints, and map them to concrete automation objectives. Handles ambiguous or incomplete specifications by asking clarifying questions or making reasonable assumptions based on app context.","intents":["I want to specify automation tasks in natural language without learning a domain-specific language","I need the agent to understand implicit constraints and context from my task description","I want the agent to handle ambiguous specifications by asking clarifying questions"],"best_for":["non-technical users specifying automation tasks","developers building user-facing automation interfaces","teams wanting to reduce automation specification overhead"],"limitations":["Intent understanding accuracy depends on task clarity; ambiguous descriptions may lead to incorrect automation","No built-in mechanism to ask clarifying questions; requires external UI for user interaction","Language understanding is limited to training data; domain-specific terminology may not be recognized","Intent extraction is lossy; some user constraints may be lost during mapping to automation objectives"],"requires":["GUI-Owl model with reasoning capability","Python 3.8+","Mobile-Agent-v3.5 framework"],"input_types":["Natural language task description","Optional: current UI screenshot for context","Optional: app domain or category"],"output_types":["Structured task specification with extracted entities and constraints","Confidence score for intent understanding","Clarifying questions if specification is ambiguous","Mapped automation objectives"],"categories":["text-generation-language","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_12","uri":"capability://memory.knowledge.action.history.tracking.and.context.management","name":"action history tracking and context management","description":"Maintains a rolling history of executed actions, screenshots, and outcomes to provide context for planning and reflection. Uses this history to detect patterns (repeated failures, circular action sequences), identify state divergence from expected trajectory, and inform replanning decisions. Implements efficient history compression to manage memory usage in long-running automations.","intents":["I want the agent to remember what actions it has already tried to avoid repeating failed attempts","I need the agent to detect when it's stuck in a loop and trigger recovery","I want to provide execution history as context for debugging failed automations"],"best_for":["long-running automation workflows that may encounter loops or deadlocks","debugging teams analyzing failed automation traces","systems needing to detect and recover from circular action sequences"],"limitations":["History storage grows linearly with automation length; long workflows may consume significant memory","History compression (summarization) may lose important details needed for debugging","Loop detection requires heuristics (e.g., repeated screenshots) which may have false positives/negatives","No built-in mechanism to learn from history across sessions; each automation starts fresh"],"requires":["Mobile-Agent-v3.5 framework","Python 3.8+","Storage for action history (in-memory or database)"],"input_types":["Executed actions with parameters","Post-action screenshots","Action outcomes (success/failure)"],"output_types":["Structured action history","Detected patterns (loops, repeated failures)","Compressed history summary for context"],"categories":["memory-knowledge","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_2","uri":"capability://automation.workflow.cross.platform.action.execution.with.unified.controller.abstraction","name":"cross-platform action execution with unified controller abstraction","description":"Provides a unified action execution layer that translates high-level GUI actions (tap, swipe, type, scroll) into platform-specific commands via pluggable controllers: AndroidController (ADB), HarmonyOSController (HarmonyOS APIs), PyAutoGUI (desktop), and Playwright (browser). Each controller implements a common interface, enabling the same action plan to execute across mobile and desktop without modification.","intents":["I want to execute the same automation workflow on Android, HarmonyOS, Windows, macOS, and web browsers","I need to translate abstract actions like 'tap element at coordinates' into platform-native commands","I want to handle platform-specific quirks (ADB latency, Playwright timeouts) transparently"],"best_for":["cross-platform automation teams supporting multiple OS targets","mobile app developers testing on both Android and HarmonyOS","QA engineers building unified test suites for web + mobile + desktop"],"limitations":["ADB-based Android execution adds 100-300ms per action due to USB/network latency; local device connection required","Playwright browser automation requires headless browser setup and may conflict with native browser security policies","PyAutoGUI desktop automation is screen-resolution dependent; coordinates must be recalibrated for different display densities","No built-in retry logic for transient failures (network drops, device disconnects); requires external orchestration"],"requires":["Platform-specific tools: Android Debug Bridge (ADB) for Android, HarmonyOS SDK for HarmonyOS, PyAutoGUI for desktop, Playwright for browsers","Python 3.8+","Mobile-Agent-v3.5 framework","Connected device or running browser instance"],"input_types":["Structured action objects: {action_type: 'tap', x: 100, y: 200}","Element references from grounding output","Optional: action parameters (text for typing, duration for swipes)"],"output_types":["Execution status (success/failure)","Post-action screenshot for state validation","Error messages with platform-specific diagnostics"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_3","uri":"capability://planning.reasoning.visual.state.validation.and.action.feedback.loop","name":"visual state validation and action feedback loop","description":"Captures post-action screenshots and uses GUI-Owl perception to validate whether the executed action achieved its intended effect (e.g., confirming a button press changed the UI state). Implements a feedback loop that detects action failures (element not clickable, network timeout) and triggers replanning or retry logic, enabling self-correcting automation without explicit error handling code.","intents":["I want the agent to verify that each action succeeded before proceeding to the next step","I need automatic detection when an action failed (e.g., click didn't register) and recovery without manual intervention","I want to understand why an action failed by analyzing the post-action UI state"],"best_for":["automation engineers building resilient, self-healing test suites","teams automating flaky mobile apps with network latency and UI race conditions","developers creating production-grade automation that must handle transient failures"],"limitations":["Screenshot capture + model inference adds 500ms-2s per action; validation latency compounds over multi-step workflows","Validation accuracy depends on model's ability to detect subtle UI state changes; may miss visual updates that don't affect element positions","No semantic understanding of 'success' — requires explicit definition of expected post-action state (e.g., 'button should be disabled')","Retry logic can create infinite loops if the UI state is fundamentally broken; requires timeout guards"],"requires":["GUI-Owl model for post-action perception","Screenshot capability on target device (ADB screencap, Playwright screenshot, PyAutoGUI)","Python 3.8+","Mobile-Agent-v3.5 or v3 framework"],"input_types":["Pre-action screenshot and action description","Post-action screenshot (captured automatically)","Optional: expected state description for validation"],"output_types":["Boolean success/failure status","Confidence score for validation","Detailed comparison of pre/post UI state","Suggested recovery action if validation fails"],"categories":["planning-reasoning","image-visual"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_4","uri":"capability://planning.reasoning.semi.online.reinforcement.learning.for.action.policy.optimization","name":"semi-online reinforcement learning for action policy optimization","description":"Implements UI-S1 training pipeline using VERL framework to fine-tune GUI-Owl models on real mobile app interactions through semi-online RL. The system collects trajectories from live app executions, generates synthetic rewards based on task completion and action efficiency, and updates the model to improve action selection without requiring manual annotation. Enables continuous improvement of automation policies as new app versions and UI patterns are encountered.","intents":["I want to improve the agent's action selection over time by learning from successful and failed executions","I need to adapt the automation policy to new app versions without manual retraining","I want to optimize for efficiency metrics like action count and execution time, not just task completion"],"best_for":["teams running continuous automation at scale who can collect large trajectory datasets","mobile app developers building self-improving test automation","organizations with resources for RL infrastructure (GPU clusters, trajectory storage)"],"limitations":["RL training requires 1000s of successful trajectories to converge; small-scale deployments may not generate sufficient data","Reward function design is critical and non-trivial; poorly designed rewards can lead to policy collapse or gaming behavior","Training adds significant computational overhead (GPU hours per update cycle); not suitable for real-time adaptation","Requires careful handling of distribution shift when app UI changes significantly; old trajectories may become invalid"],"requires":["VERL framework for RL training","GPU cluster (8+ GPUs recommended for reasonable training speed)","Python 3.8+","Large trajectory dataset from live app executions (1000+ successful trajectories minimum)","Mobile-Agent-v3.5 framework with training utilities"],"input_types":["Trajectory data: sequence of (screenshot, action, reward) tuples","Task descriptions and success criteria","Optional: human feedback on action quality"],"output_types":["Fine-tuned GUI-Owl model weights","Training metrics (reward curves, action distribution shifts)","Policy evaluation results on held-out test set"],"categories":["planning-reasoning","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_5","uri":"capability://planning.reasoning.pre.operative.error.diagnosis.with.gui.critic.r1","name":"pre-operative error diagnosis with gui-critic-r1","description":"Implements GUI-Critic-R1 module that analyzes planned action sequences before execution to predict and diagnose potential failures (unreachable elements, invalid state transitions, missing prerequisites). Uses extended reasoning to evaluate action feasibility against current UI state and generates diagnostic reports with suggested corrections, reducing failed executions and improving overall automation reliability.","intents":["I want to catch action plan errors before they execute and waste time","I need to understand why a planned action might fail (e.g., element not visible, wrong state)","I want suggestions for correcting invalid action sequences"],"best_for":["automation teams running expensive mobile app tests where failures are costly","developers building production automation that must maintain high success rates","QA engineers debugging complex multi-step workflows"],"limitations":["Diagnostic accuracy depends on model's ability to reason about UI state; may miss edge cases or race conditions","Pre-operative analysis adds latency (1-3 seconds per action sequence) before execution begins","Cannot detect failures caused by external factors (network timeouts, device crashes) that occur during execution","Requires explicit definition of valid state transitions; implicit assumptions about UI behavior may not be captured"],"requires":["GUI-Owl model with extended reasoning capability (Thinking variant)","Python 3.8+","Mobile-Agent-v3.5 framework","Current UI screenshot for context"],"input_types":["Planned action sequence with target elements and parameters","Current UI screenshot","Optional: app state constraints and valid transitions"],"output_types":["Feasibility score for each action (0-1)","Diagnostic report with identified issues","Suggested corrections or alternative action sequences","Confidence in diagnosis"],"categories":["planning-reasoning","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_6","uri":"capability://automation.workflow.desktop.and.browser.automation.with.platform.specific.controllers","name":"desktop and browser automation with platform-specific controllers","description":"Extends Mobile-Agent framework to desktop (Windows/macOS/Linux) and web browsers through PC-Agent and Playwright-based controllers. Implements platform-specific element detection (Windows UI Automation, macOS Accessibility APIs, DOM parsing for web) and action execution (pywinauto, macOS native APIs, Playwright commands), enabling unified automation across mobile, desktop, and web with minimal code changes.","intents":["I want to automate desktop applications (Windows, macOS, Linux) using the same agent framework as mobile","I need to automate web applications in headless browsers without manual Selenium/Playwright code","I want to test cross-platform workflows that span mobile, desktop, and web"],"best_for":["QA teams testing desktop + mobile + web applications","automation engineers building unified test suites across platforms","developers creating cross-platform RPA solutions"],"limitations":["Desktop automation (pywinauto, macOS APIs) requires elevated privileges and may conflict with system security policies","Playwright browser automation is limited to headless/headed browser contexts; native browser extensions and plugins may not work","Element detection on desktop relies on platform-specific APIs (UI Automation, Accessibility) which have different capabilities and performance characteristics","Screen resolution and DPI scaling affect coordinate-based actions; requires calibration per display configuration"],"requires":["PC-Agent framework for desktop automation","Platform-specific tools: pywinauto (Windows), macOS Accessibility APIs (macOS), Playwright (browsers)","Python 3.8+","GUI-Owl model for visual understanding","Elevated privileges for desktop automation (admin/sudo)"],"input_types":["Natural language task description","Desktop/browser screenshots","Optional: DOM tree for web applications"],"output_types":["Execution status and action trace","Post-action screenshots","Error logs with platform-specific diagnostics"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_7","uri":"capability://planning.reasoning.self.evolving.agent.with.continuous.capability.expansion","name":"self-evolving agent with continuous capability expansion","description":"Mobile-Agent-E implements self-evolution mechanism where the agent learns new capabilities and refines existing ones through interaction with diverse apps and user feedback. The system maintains a capability registry, collects execution traces, and uses reinforcement learning to expand the action vocabulary and improve decision-making for novel UI patterns not seen during initial training.","intents":["I want the agent to handle new app types and UI patterns without retraining from scratch","I need the agent to learn from user corrections and feedback to improve future executions","I want continuous improvement of automation capabilities as the agent encounters new scenarios"],"best_for":["organizations running long-lived automation systems that encounter diverse apps","teams with continuous feedback loops (user corrections, execution logs)","developers building adaptive automation that improves over time"],"limitations":["Self-evolution requires careful safeguards to prevent learning from incorrect feedback; malicious or noisy corrections can degrade performance","Capability expansion is slow and requires many interactions to converge; not suitable for rapid deployment scenarios","No built-in mechanism to unlearn outdated capabilities when app UIs change; requires manual capability pruning","Requires persistent storage of capability registry and execution traces; adds infrastructure complexity"],"requires":["Mobile-Agent-E framework","Python 3.8+","Persistent storage for capability registry (database or file system)","User feedback collection mechanism","GPU for continuous fine-tuning"],"input_types":["Execution traces (screenshots, actions, outcomes)","User feedback on action quality","New app screenshots and task descriptions"],"output_types":["Updated capability registry","Fine-tuned model weights","Capability improvement metrics"],"categories":["planning-reasoning","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_8","uri":"capability://planning.reasoning.multi.agent.orchestration.and.task.delegation","name":"multi-agent orchestration and task delegation","description":"Mobile-Agent-v2 implements multi-agent system where specialized agents handle different aspects of automation: planning agent decomposes tasks, execution agent performs actions, reflection agent validates outcomes and triggers replanning. Agents communicate through shared state (screenshots, action history) and coordinate via a central orchestrator that manages task flow and error recovery.","intents":["I want to decompose complex automation into specialized agents with different responsibilities","I need robust error recovery where failed actions trigger reflection and replanning","I want to parallelize independent subtasks across multiple agents"],"best_for":["teams automating complex, multi-stage workflows with independent subtasks","organizations needing robust error recovery and replanning capabilities","developers building modular automation systems with clear separation of concerns"],"limitations":["Multi-agent coordination adds latency due to inter-agent communication and state synchronization","Shared state management (screenshots, action history) requires careful synchronization to prevent race conditions","Agent specialization requires careful prompt engineering and role definition; poorly designed agents can create bottlenecks","Debugging multi-agent systems is complex; failures can originate from any agent or coordination layer"],"requires":["Mobile-Agent-v2 framework","Python 3.8+","LLM API access (GPT-4o in original implementation) or local GUI-Owl models","Shared state storage (in-memory or database)"],"input_types":["High-level task description","Current UI screenshot","Optional: task decomposition hints"],"output_types":["Execution trace with per-agent actions","Final outcome and success status","Reflection report on failures and recovery actions"],"categories":["planning-reasoning","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"github-x-plug--mobileagent__cap_9","uri":"capability://memory.knowledge.knowledge.base.and.gui.element.semantic.understanding","name":"knowledge base and gui element semantic understanding","description":"Maintains a knowledge base of common UI patterns, element types, and interaction semantics across diverse apps. GUI-Owl models leverage this knowledge during perception and planning to understand element purpose (button, input field, navigation) and predict likely interactions, improving grounding accuracy and action selection without requiring app-specific training.","intents":["I want the agent to understand common UI patterns and element types without app-specific training","I need semantic understanding of element purpose to predict valid interactions","I want to leverage knowledge of similar apps to improve automation on new apps"],"best_for":["automation teams working with diverse apps that share common UI patterns","developers building general-purpose mobile automation without app-specific customization","organizations wanting to reduce per-app training overhead"],"limitations":["Knowledge base coverage is limited to patterns seen during training; novel UI patterns may not be recognized","Semantic understanding can be incorrect for non-standard element usage (e.g., button used as text display); requires validation","Knowledge base updates require retraining or fine-tuning; static knowledge becomes stale as UI patterns evolve","Cross-app knowledge transfer may introduce false assumptions about interaction semantics"],"requires":["GUI-Owl model with knowledge base integration","Python 3.8+","Mobile-Agent-v3.5 framework","Optional: custom knowledge base for domain-specific apps"],"input_types":["UI screenshots","Element descriptions and context","Optional: app domain or category"],"output_types":["Element semantic labels (button, input, navigation, etc.)","Predicted interaction types","Confidence scores for semantic understanding"],"categories":["memory-knowledge","image-visual"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":47,"verified":false,"data_access_risk":"high","permissions":["GUI-Owl model weights (1.5, 7B, or 32B variant) from Alibaba/Tongyi Lab","Python 3.8+","CUDA 11.8+ for GPU acceleration (optional but recommended)","Mobile device or desktop with ADB/Playwright/PyAutoGUI connectivity for screenshot capture","GUI-Owl model with Thinking capability enabled","Mobile-Agent-v3.5 or v3 framework","Connected Android device or desktop environment for action execution","Mobile-Agent-v3.5 framework with evaluation utilities","Connected Android device or emulator","Benchmark app APKs (provided in repository)"],"failure_modes":["Model inference latency varies by variant (1.5 faster than 7B/32B); larger models provide better reasoning but slower grounding","Grounding accuracy degrades on heavily obfuscated or custom-rendered UI elements not well-represented in training data","Requires GPU for reasonable inference speed; CPU-only inference adds 2-5x latency overhead","Planning latency increases with task complexity; 5-10 step tasks typically require 2-5 seconds of model inference","Reasoning quality depends on model variant; GUI-Owl-1.5 provides faster planning but GUI-Owl-32B offers more robust decomposition for ambiguous intents","No built-in persistent memory across sessions; each task planning starts fresh without learning from previous executions","Benchmarks may not reflect real-world app diversity; performance on benchmarks doesn't guarantee performance on unseen apps","Evaluation requires running full automation workflows which is time-consuming; benchmark suites may take hours to complete","Metrics like 'task completion' require manual definition of success criteria; different definitions lead to incomparable results","Benchmark apps may change over time (app updates, UI redesigns) making historical comparisons invalid","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.6457097747619172,"quality":0.35,"ecosystem":0.6000000000000001,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.25,"quality":0.25,"ecosystem":0.1,"match_graph":0.28,"freshness":0.12}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"active","updated_at":"2026-05-24T12:16:22.064Z","last_scraped_at":"2026-05-03T13:58:39.623Z","last_commit":"2026-04-14T08:30:36Z"},"community":{"stars":8609,"forks":870,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=x-plug--mobileagent","compare_url":"https://unfragile.ai/compare?artifact=x-plug--mobileagent"}},"signature":"kTc5j+DP9CeCnw/cUlv+uka5jXHvbsRVxcbNqVomRQc1xt4OV4NvHDgYAjywDkNmH2bLFiZ1YzzWXnmUpuvdBA==","signedAt":"2026-06-20T03:55:59.408Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/x-plug--mobileagent","artifact":"https://unfragile.ai/x-plug--mobileagent","verify":"https://unfragile.ai/api/v1/verify?slug=x-plug--mobileagent","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}