RoboClaw Co-Pilot · BYOK AI Router · v3.16

From Manual Test Cases to
Production-Ready Scripts
in Minutes.

RoboClaw generates complete Robot Framework automation — .robot files, keywords, locators, and Python libraries — grounded in your actual codebase. Dedicated Co-Pilot chat, BYOK AI router (Claude, OpenAI, Gemini, Azure, Ollama…), drag-and-drop RAG training data, Live Device Mirror, fingerprint-based self-healing, 11 MCP servers, and dual ADO / Jira support. No hallucinations. No lock-in.

See How It Works
Built for
Robot Framework · Appium · Selenium · VS Code · MCP · Claude · OpenAI · Gemini · Copilot · Azure DevOps · Jira
RoboClaw v2 — Processing Engine
$ roboclaw pipeline --test-case US-20609625 --live
AI Router → Anthropic · claude-3-5-sonnet (BYOK)
Live Device Mirror — streaming iPad Mini @ 12 FPS...
RAG training data attached — 6 files, 24.8 KB context
Loop 1 RAG → Anti-Hallucination → Reuse → Locator → Self-Heal
24/24 locators verified · fingerprinted (SHA-256)
Output: MainTestRobot.robot + 3 PageRobots + CustomLib.py
💬 RoboClaw Co-Pilot → “Explain the retest failure”
CodyView review — 18/18 quality gates green
robot --include smoke Tests/ → 12 pass · 0 fail
Self-healing watcher armed on output.xml · ready to heal.
🛡 Fallback: Copilot LM available · offline template ready
🔒 Security scan passed — 0 OWASP findings · secrets redacted
Scroll

Up and Running in
60 Seconds

# After downloading the .vsix, open terminal in the download folder:
$ code --install-extension roboclaw-ai-studio-3.16.7.vsix
Watch Demo
10xFaster
95%Accuracy
0Hallucinations
9AI Providers
11MCP Servers
10Healing Strategies

RoboClaw AI Studio
Inside Your IDE

A guided, single-pane workflow built right into VS Code — stream your device in real time, import a test case, capture screens, and generate production-ready Robot Framework automation — then keep it green with fingerprint-based self-healing.

VS Code · RoboClaw AI Studio · Capture Studio
Live
RoboClaw AI Studio inside VS Code — Capture Studio sidebar showing connected iPad Mini, imported ADO test case, workflow modes (TC Gen, Live App, Design, Builder, CI/CD), and step-by-step quick actions next to a generated Robot Framework test file
Live device connected
iPad Mini · iOS 2025.8
12/12 quality gates
0 hallucinated locators
5-loop generation
RAG-grounded · < 3 min
📱

Live Device Mirror

Real-time JPEG stream with pointer, swipe & keyboard passthrough — plus element inspector overlays.

📋

Import from ADO or Jira

Bring test cases straight from Azure DevOps, Jira, CSV, or Markdown — parsed and ready for capture.

📸

Capture, Annotate, Generate

Single capture button fetches screenshot + XML, annotates elements, and feeds the 5-loop engine.

🩺

Self-Healing Watcher

Auto-analyses output.xml after every run — heals locators above 0.80 confidence, queues the rest for review.

Manual Test Script Creation
Is a Bottleneck

⏱️

Hours of Manual Work

Writing .robot files, building XPath locators, managing page objects — all by hand, screen by screen.

🔄

Duplicated Effort

Same keywords and locators rewritten across test cases. No reuse intelligence. No workspace awareness.

🎭

AI Hallucinations

Generic AI tools invent locators, fabricate keywords, and produce scripts that fail on first run.

🔧

Fragile Locators

Hard-coded XPaths break with every UI update. No healing, no fallback strategies, constant maintenance.

Everything You Need to
Ship Automation Faster

RoboClaw's 5-loop Processing Engine turns test cases into verified, production-ready automation scripts — grounded in your workspace, not hallucinated from thin air.

Core Engine

RAG-Grounded Generation

Every keyword, variable, and locator is retrieved from your actual codebase before generation. RoboClaw searches existing .robot files, PageRobots, and Python libraries — reusing what exists and only creating what's genuinely new.

14 keywords reused → 3 new created → 0 hallucinated

Live Device Mirror

Real-time JPEG stream of any connected iOS, Android, or Web target with pointer, swipe & keyboard passthrough. Element inspector overlays bounding boxes, adjustable FPS / quality / width, beside-editor pop-out.

Anti-Hallucination Engine

Every generated locator is validated against real Appium XML page sources. If it doesn't exist in the DOM, it doesn't ship.

5-Tier XPath Classification

Static, dynamic, qualified, indexed, and iOS class chain — every element gets the most resilient locator strategy automatically.

Fingerprint Self-Healing

SHA-256 element fingerprints + 10 healing strategies with confidence scoring. Auto-heals at ≥0.80, queues 0.5–0.8 for review, escalates the rest to Copilot. Watches output.xml live.

Multi-Source Input

ADO test cases, Jira stories, CSV files, screenshots, XML page sources, TOON annotation reports — feed any input and get complete automation output.

12-Point Quality Gates

Every generated file passes a comprehensive validation checklist — file structure, keyword deduplication, locator accuracy, import verification, and more.

Screenshot Annotation

Color-coded bounding boxes with visual legends and structured TOON reports. Every element annotated with XPath, accessibility ID, and iOS class chain.

VS Code Extension

RoboClaw AI Studio — a guided workflow panel inside VS Code that automates the entire capture-to-generation pipeline with single-click actions.

Test Case Generation

AI-powered test case authoring from user stories and acceptance criteria — producing structured, priority-classified test cases with full boundary and negative coverage.

API Test Automation

End-to-end API validation suites using RequestsLibrary — automated request construction, response assertion, schema validation, and chained workflow testing.

Design Phase Skeleton Generator

Import Figma designs and test cases during the design phase to generate production-format Robot Framework skeletons — Main Test File with DataDriver + Test Template and PageRobots with <<<PLACEHOLDER>>> locators ready for real values.

Dynamic Appium Capabilities

Automatic device detection and capability management — pass UDID, device name, platform version, and automation name dynamically. No more hardcoded device configs breaking teammate setups.

Security Hardened

OWASP Top 10 compliant — XSS protection, injection prevention, secret redaction, path traversal guards, HTTPS enforcement, and URL validation built into every integration point.

Setup Wizard

First-launch guided setup with name, license code, and project management configuration. Choose between Azure DevOps or Atlassian Jira — the entire UI adapts dynamically to your selection.

Azure DevOps + Jira Integration

Dual project management support — import user stories, test cases, and deploy results to either Azure DevOps or Atlassian Jira. Switch seamlessly via Setup Wizard or Settings.

TC Generation Mode

AI-powered test case generation in three modes — Functional, Regression, and Smoke. Import user stories and acceptance criteria, generate structured test cases with boundary, negative, and UI coverage.

Rules Engine (Do's & Don'ts)

Define generation rules through a visual modal — specify coding standards, naming conventions, and anti-patterns. Rules are saved as Markdown and auto-attached to every Copilot invocation.

Normal & Factory Reset

Two-tier reset system — Normal Reset clears the current session, while Factory Reset wipes all stored credentials, settings, and setup data, then re-launches the Setup Wizard fresh.

Co-Pilot Chat, BYOK Router,
MCP & Bug Filing

RoboClaw goes beyond script generation — chat with a dedicated Co-Pilot, route AI through any provider you bring, drag-and-drop training data, stream devices live, file bugs from failures, load-test with HAR captures, and expose every capability through 11 standalone MCP servers.

New in v3.16

RoboClaw Co-Pilot

Dedicated in-editor chat surface — streaming responses, code-block copy, per-turn regenerate, attach-file for RAG context, and one-tap quick actions: Generate Bug Report, Create Smoke TC, Generate Robot Script, Explain This Failure. Pop out to a browser PWA to share transcripts.

Beside-editor WebviewPanel · status-bar launcher · floating chip · last-50-turn history
BYOK Router

Bring-Your-Own-Key AI Router

One central aiRouter dispatches every AI call — BugBuddy, Retest, Self-Heal, Exploratory, TC Gen, Co-Pilot — through your choice of provider. GitHub Copilot is always available as a fallback. Live-validated keys, per-provider model picker, and a Route via stamp on every generated artifact.

Anthropic · OpenAI · Google Gemini · Azure OpenAI · Ollama · LM Studio · OpenRouter · Groq · GitHub Copilot

RAG Training Data Icon

New sidebar header icon opens a drag-and-drop modal for .md, .json, .txt, .docx, .pdf files. Every AI call automatically injects up to 30 KB of training context — so your project conventions, glossaries, and prior tickets ground every response. Live count badge on the icon.

Copilot Fallback — Always On

Even without a BYOK key, RoboClaw runs. Every AI feature transparently falls back to GitHub Copilot's Language Model API, then to a deterministic offline template if Copilot is unavailable. Zero-config path from install to first generated script.

MCP Suite

11 MCP Servers

Every RoboClaw capability exposed as a Model Context Protocol server — plug into Claude Desktop, VS Code Copilot Chat, Cursor, or Continue. FastMCP-based, stdio transport, fully composable.

robot-framework · locator-engine · appium-session · workspace-rag · azure-devops · screenshot-annotator · test-data-generator · page-object · healing-engine · config-robot · cicd-pipeline

BugBuddy

One-click bug filing from failures, selections, or test conclusions. Files structured Bugs in Azure DevOps or Jira with steps-to-reproduce, severity, screenshots, & custom fields. Retest comments, history view, & configurable field mappings.

Performance Testing

Record a scenario or import HAR files, run load tests with configurable thresholds, watch results live on the dashboard, compare runs, and export PDF reports. Built for Robot Framework + custom scenarios.

Exploratory Session

Open a guided exploratory testing session — capture screens on demand, get AI-suggested edge cases, pause & resume, and convert findings into structured test cases. Pure investigation, zero scripting overhead.

XCUITest ↔ UiAutomator2

One-click conversion between iOS XCUITest and Android UiAutomator2 locator syntax — reuse the same test logic across platforms without rewriting selectors by hand.

Workflow Builder

Load any captured test case and visually map every step to a screenshot + locator. Export the workflow mapping as JSON, jump from any step into its source screen, and hand off to the generator with full visual context.

BYOK Claude API

Bring Your Own Key — your Claude API key is stored in VS Code SecretStorage, never logged, never sent anywhere except Anthropic. No RoboClaw backend. Everything else runs locally.

output.xml Watcher

Detects Robot Framework runs the moment they finish, parses output.xml, fingerprint-matches failed locators against the healing DB, and triggers immediate re-heal attempts — zero manual action required.

Run Separate & Dry-Run by Tag

Multi-pick test files and run them with isolated output dirs, or dry-run by tag (--include smoke) to validate suites before they hit a real device.

Documentation
for Every Workflow

Focused, screenshot-rich guides covering install, manual test-case generation, BugBuddy, live device mirroring, and the self-healing system — each linked from inside the extension too.

The 5-Loop
Intelligence Pipeline

Every test case passes through five sequential intelligence loops before a single line of code is generated.

01

RAG Retrieval

Scans your entire workspace — .robot files, PageRobots, Python libraries, variables — building a context map of everything that already exists.

02

Anti-Hallucination

Cross-references every generated element against real Appium XML page sources. Zero tolerance for fabricated locators or invented keywords.

03

Reusability Logic

Intelligent deduplication. If a keyword exists in PageRobots, it's reused. If similar logic exists, it's adapted. Only truly new functionality gets new code.

04

Locator Mapping

5-tier XPath classification runs the deterministic Locator Engine — producing static, dynamic, qualified, indexed, and class chain strategies for every element.

05

Self-Learning

Every generation cycle feeds back into RoboClaw's knowledge base. Healing events, new patterns, and validation results continuously improve future output.

A Fleet of Specialized
AI Sub-Agents

RoboClaw orchestrates a team of focused agents, each responsible for a critical phase of the automation lifecycle.

Ⓜ️

RoboClaw

Orchestrator

The core agent. Processes test cases through the 5-loop engine, coordinates sub-agents, and produces complete automation artifacts.

🅢

CodyView

Code Reviewer

18-point quality checklist, RAG consistency cross-check, anti-pattern detection, duplicate keyword scanning, and XPath validation.

🅢

StepAnalyzer

Pre-Generation Intel

Reads CSV test cases and classifies each step as REUSE, SUGGEST, REFERENCE, or NEW — before any code is written.

🅢

ScreenshotAnnotator

Visual Intelligence

Annotates screenshots with color-coded bounding boxes. Produces TOON reports with XPaths, accessibility IDs, and locator strategies.

🅢

PostProcessor

Quality Gate

Human-in-the-loop review. Flags ambiguous mappings, low-confidence locators, and new patterns for confirmation before finalizing.

🅢

SelfHealingAgent

Runtime Healer

Monitors execution logs for locator failures. 10 healing strategies with confidence scoring. SQLite-backed healing event tracking.

🅢

DQ Agent

Domain Specialist

Digital-Quality line/category iteration agent — generates and orchestrates Robot suites for repetitive manufacturing & quality workflows.

One Agent.
Every Platform.

Mobile

iOS & Android via Appium. Native, hybrid, and Flutter apps with Appium Device Farm support for parallel execution.

Appium XCUITest UIAutomator2

Web

Chrome, Firefox, Safari, Edge with Selenium and Browser Library. Headless mode for CI/CD pipelines.

Selenium Browser Library Playwright

API

REST API testing with RequestsLibrary. JSON schema validation, bearer token management, and API + Mobile hybrid tests.

RequestsLibrary JSON REST

Database

Direct database validation with DatabaseLibrary. Verify data integrity, stored procedures, and end-to-end data flows.

DatabaseLibrary SQL Oracle

Numbers That
Speak for Themselves

90%
Reduction in Script Creation Time
From 30–60 min to 3–5 min per test case
0%
Hallucinated Locators
Every XPath validated against real page sources
50
Commands
Capture, locator, healing, TC Gen, BugBuddy, perf, exploratory & more
11
MCP Servers
FastMCP · stdio · Claude Desktop, VS Code, Cursor, Continue
7
Workflow Steps
Configure → Import → Capture → Locators → Analyzer → Heal → Run
2
PM Integrations
Azure DevOps + Atlassian Jira — import, generate & deploy
10
Healing Strategies
SHA-256 fingerprint + 10 fallbacks with confidence scoring
5
Processing Loops
RAG → Anti-Hallucination → Reuse → Locators → Learn
7
Specialized Agents
Orchestrator + 6 sub-agents working in concert
0
Security Vulnerabilities
OWASP Top 10 compliant — XSS, injection & traversal hardened

What RoboClaw
Actually Generates

MainTestRobot.robot Test File
*** Settings ***
Library    DataDriver    file=TestData.csv
Resource   ../PageRobots/MyShift_PageRobot.robot
Resource   ../PageRobots/ConfirmShift_PageRobot.robot
Resource   ../Resources/GenericLib/GlobalSetting.robot
Test Template    Verify Confirm Shift Flow

*** Test Cases ***
TC_${TestID}    # DataDriver auto-generates

*** Keywords ***
Verify Confirm Shift Flow
    [Arguments]    ${TestID}    ${Line}    ${Expected}
    Navigate To My Shift Screen
    Select Manufacturing Line    ${Line}
    Tap Confirm Shift Button
    Verify Confirmation Status    ${Expected}
MyShift_PageRobot.robot PageRobot
*** Settings ***
Library    AppiumLibrary
Library    ../Libraries/MobileLib.py
Resource   ../Resources/GenericLib/GlobalSetting.robot

*** Variables ***
${MY_SHIFT_HEADER}    xpath=//XCUIElementTypeStaticText[@name="My Shift"]
${CONFIRM_BTN}       xpath=//XCUIElementTypeButton[@name="Confirm"]
${LINE_CHIP}         xpath=//XCUIElementTypeCell[contains(@name,"Line")]

*** Keywords ***
Navigate To My Shift Screen
    Wait Until Element Is Visible    ${MY_SHIFT_HEADER}    ${TIMEOUT}
    Element Should Be Visible    ${MY_SHIFT_HEADER}

Tap Confirm Shift Button
    Wait Until Element Is Visible    ${CONFIRM_BTN}    ${TIMEOUT}
    Click Element    ${CONFIRM_BTN}

Deterministic Pipeline.
Intelligent Generation.

⚙️
1 · Configure Session
📥
2 · Import TC
📸
3 · Capture / Live
🔍
4 · Locators & Annotate
🧪
5 · Analyzer / Debug
🩺
6 · Self-Healing
▶️
7 · Run & Heal

Live Device Mirror

NDJSON-framed JPEG stream from Python helper → inspector overlay → pointer/swipe/keyboard passthrough to Appium / Selenium

Locator Engine

5-tier XPath + iOS Predicate + Class Chain + CSS + Accessibility ID — confidence-scored alternatives per element

RAG Generation

5-loop pipeline: Retrieval → Anti-hallucination → REUSE/SUGGEST/NEW → Locator binding → Self-Learning

Self-Healing Loop

SHA-256 fingerprints + 10 strategies + output.xml watcher → auto-heal ≥0.80, queue 0.5–0.8 for review, Copilot fallback below

Ready to Eliminate Manual
Test Script Creation?

Join the teams using RoboClaw to ship automation 10x faster — with zero hallucinations and self-healing locators.

After downloading, install the .vsix in one command:
code --install-extension roboclaw-ai-studio-3.16.7.vsix
1 Download the .vsix file using the button above
2 Open your workspace, run the Setup Wizard (name · license · ADO or Jira) — all 40+ Python deps install offline
3 Click Configure Session, start the Live Device Mirror, and run the 7-step workflow end-to-end

The Story Behind
RoboClaw AI Studio

Born from real-world frustration with manual test automation — engineered to eliminate every bottleneck in the QA pipeline.

Origin

From 60-Minute Scripts to 3-Minute Generation

RoboClaw AI Studio was created by Abdul Talib — a senior QA automation engineer who spent years watching teams waste hundreds of hours writing repetitive Robot Framework scripts by hand. Every new test case meant the same tedious cycle: manually inspecting Appium XML, crafting XPath locators, writing keywords, managing page objects, and debugging scripts that broke whenever the UI changed.

The breaking point came during a large-scale MFG Pro+ iPad application project where the team needed to automate hundreds of test cases across complex manufacturing workflows — Confirm Shift, Digital Quality, Line Performance, and more. Manual scripting couldn't keep pace with the sprint velocity. Something fundamentally different was needed.

That's when RoboClaw was born — not as another code generation tool, but as a RAG-enhanced AI agent that understands your existing codebase, validates every output against real page sources, and produces scripts that actually run on the first attempt.

Sept 2025 Initial concept — Python scripts for locator extraction
Oct 2025 v1.0 — First VS Code extension with Appium capture
Jan 2026 v2.0 — RAG engine, 5-loop pipeline, sub-agents
Mar 2026 v3.0 — TC Gen, Design mode, ADO + Jira, Workflow Builder
Apr 2026 v3.11 — CI/CD Pipeline, exploratory session, security hardening
May 2026 v3.13 — Live Device Mirror (Phase 1–4), Dynamic XPath builder
May 2026 v3.14 — Fingerprint Self-Healing engine + dashboard + output.xml watcher
Jun 2026 v3.15 — 11 MCP servers, BugBuddy, Performance testing, 7-step workflow polish
Jul 2026 v3.16 — RoboClaw AI Studio rebrand · BYOK AI Router · Co-Pilot chat · RAG training data icon · Copilot fallback

Mission

Eliminate the gap between test design and test automation. Every test case written by a QA analyst should become a production-ready automation script — instantly, accurately, and without a single hallucinated locator.

Vision

A world where QA engineers focus on test strategy and exploratory testing — while AI handles the mechanical work of script creation, locator management, self-healing, and CI/CD pipeline integration.

Core Principles

Zero hallucinations. Workspace-grounded intelligence. Deterministic output. Self-healing resilience. Security by default. Every design decision is driven by these non-negotiable principles.

What Makes RoboClaw Fundamentally Different

Generic AI
RoboClaw

Workspace Awareness

Generic AI tools generate code in a vacuum. RoboClaw scans your entire workspace — every .robot file, every PageRobot, every Python library — before generating a single line. It knows what already exists and reuses it.

Generic AI
RoboClaw

Locator Verification

ChatGPT and Copilot invent XPath locators that look plausible but don't exist in the DOM. RoboClaw validates every locator against real Appium XML page sources — if it's not in the DOM, it doesn't ship.

Generic AI
RoboClaw

Production Architecture

Generic tools dump everything in one file. RoboClaw produces the correct architecture: Main Test File (DataDriver + Test Template), separate PageRobots per component, shared variable files, and custom Python libraries.

Generic AI
RoboClaw

Self-Healing + Learning

Static scripts break on every UI change. RoboClaw's self-healing engine monitors execution, auto-repairs locators with 10 strategies, tracks confidence scores, and feeds results back into future generation.

Technology Under the Hood

RAG System

Lightweight JSON-based Retrieval-Augmented Generation that indexes your workspace keywords, variables, locators, and libraries in real-time.

VS Code Extension API

WebviewView providers (sidebar + dashboard + live mirror pop-out), 50+ commands, status bar, NDJSON-framed long-lived Python helpers, SecretStorage for all credentials.

Locator Engine

Python-based deterministic XPath generator with 5-tier classification — static, dynamic, qualified, indexed, and iOS class chain strategies.

Fingerprint Self-Healing

SHA-256 element identity + 10 strategies (attribute fuzzy, parent/ancestor, sibling-relative, text content, structural, similarity, CSS, accessibility-id, ML-predicted, visual template) with auto-heal & manual-review thresholds.

Robot Framework

DataDriver + Test Template architecture, AppiumLibrary, SeleniumLibrary, RequestsLibrary, DatabaseLibrary — full cross-platform output.

Security Layer

OWASP Top 10 hardened — XSS sanitization, injection prevention, secret redaction, URL validation, HTTPS enforcement at every integration point.

CI/CD Pipeline Engine

15 built-in tools across 4 tiers — ADO Pipelines, GitHub Actions, Docker, pabot, Slack/Teams notifications, coverage gates, and artifact archival.

11 MCP Servers

FastMCP-based, stdio transport — robot-framework, locator-engine, appium-session, workspace-rag, azure-devops, screenshot-annotator, test-data-generator, page-object, healing-engine, config-robot, cicd-pipeline.

RoboClaw By the Numbers

3.16.7
Current Version
RoboClaw rebrand · Co-Pilot chat · BYOK router · RAG training data
5
Processing Loops
RAG, Anti-Hallucination, Reuse, Locators, Learning
7
AI Agents
Orchestrator + 6 specialized sub-agents (incl. DQ Agent)
15
CI/CD Tools
4-tier pipeline: Quick Wins → DevOps Ecosystem
10
Healing Strategies
Attribute, sibling, parent, text, accessibility & more
18
Quality Checkpoints
CodyView's automated code review checklist
5
XPath Tiers
Static, dynamic, qualified, indexed, class chain
4
Platforms
Mobile, Web, API, Database — one agent
5
Operating Modes
TC Gen, Live App, Design, Builder, CI/CD
11
MCP Servers
FastMCP suite — Claude, Copilot, Cursor, Continue ready
7
Workflow Steps
Configure → Import → Capture → Locators → Analyzer → Heal → Run
40+
Auto-Installed Deps
Fully offline bundle — zero manual pip installs
2
PM Systems
Azure DevOps + Atlassian Jira — dual integration
0
Hallucinated Locators
Every XPath validated against real page source XML
AT

Abdul Talib

SDET & AI Product Developer

SDET & AI Product Developer specializing in mobile and enterprise test automation. Built RoboClaw AI Studio from scratch to solve the real bottlenecks faced by QA teams — turning months of manual scripting into minutes of AI-powered generation. Every feature in RoboClaw was born from hands-on experience automating complex manufacturing, supply chain, and enterprise applications.

Robot Framework Appium AI/RAG VS Code Extensions Python CI/CD