Multi-Engine RPA Skills Platform for Anti-Bot, Reverse-Engineering & Mobile Automation
The RPA Agent Skills System abstracts browser, mobile, JS reverse-engineering, anti-bot, and workflow-orchestration capabilities into a standardized skill set (Skill + MCP dual-shell) that AI Agents can call directly. It integrates open-source engines such as CloakBrowser / Obscura / BrowserCluster / AgentGo / Lamda, hiding multi-engine fragmentation behind a unified interface, so Agents can treat "automation" as composable, reusable tools even against real-world scenarios like strong anti-bot, encrypted params, and mobile control.
Each project is annotated with its position in the product system, making it easy to see at a glance how systems support each other.
Upgrade "RPA automation" from scattered scripts into a standardized skills platform that Agents can schedule.
Business logic is written once (skills_impl); both the Claude Skill CLI and the single-process MCP are thin wrappers, avoiding dual maintenance and behavior drift.
Converge the heterogeneous APIs of CloakBrowser / Obscura / BrowserCluster / AgentGo / Lamda into one interface; Agents need not care about the underlying engine.
Browser skills run locally against the engine (direct) or submit tasks to BrowserCluster for remote Worker execution (cluster) — handy in dev, scalable in prod.
Each capability is exposed as a namespaced tool (browser.* / proxy.* / data.* / antibot.*); Agents compose RPA flows like calling functions.
In real automation, the hardest part is never "click once" — it is bypassing defenses, reproducing encryption, and holding sessions.
Akamai BMP / Cloudflare Turnstile / reCAPTCHA v3 / DataDome / Kasada block naive scripts instantly. anti-bot-defender generates adversarial cookies, solves challenges, and reports fingerprint risks.
Request params are encrypted and signing algorithms hide in obfuscated JS, hard to reproduce stably. web-reverser systematizes via Observe→Hook→Breakpoint→Trace→Extract→Rebuild.
Android devices are scattered, cloud-phones/emulators are complex to onboard, UI automation is fragile. mobile-controller offers 160+ APIs via Lamda for discovery, UI automation, remote desktop.
Login state is hard to persist, cross-session reuse is hard, a fingerprint change triggers risk control. session-manager snapshots/restores Cookie & LocalStorage, with CloakBrowser persistent Profile.
High ban rate, geo mismatch triggers risk control, slow responses hurt crawling. proxy-pool-manager offers 4 strategies: round-robin / least-failed / geo-match / fastest.
Hardcoded keys, plaintext logs, scattered credentials hard to audit. security-guardian does static credential audit, SecureString validation, log masking, audit trail.
RPA flows are hand-wired, non-visual, hard to reuse. workflow-orchestrator orchestrates via DAG; task-scheduler takes over cron/interval execution.
Each open-source engine has a different API, high switching cost, capabilities can't compose. Engine Layer unified abstraction + Skill Layer composition make capabilities Lego-like.
A skill matrix spanning browser, mobile, reverse-engineering, anti-bot, orchestration, and security, ready for Agents to orchestrate.
Browser lifecycle management, direct/cluster dual-mode: launch / navigate / click / type / screenshot / execute_script.
Multi-mode structured extraction: XPath / CSS / GNE / LLM, covering news sites and SPA API interception.
Counter Akamai / Cloudflare / reCAPTCHA / DataDome / Kasada: generate adversarial cookies, solve challenges, report fingerprint risks.
JS reverse-engineering workflow: page observation, hook sampling, breakpoint debugging, network analysis, local rebuild.
Android automation: device discovery, UI automation, OCR, remote desktop — 160+ APIs via Lamda.
DAG visual flow orchestration, forwarded to BrowserCluster Vue Flow backend — no self-built DAG engine.
Proxy IP lifecycle management, 4 allocation strategies, SQLite persistence and stats.
Multi-platform session snapshot/restore/import-export: Browser / Desktop / Mobile session types.
Static credential audit, SecureString validation, sensitive-data encryption and audit trail.
Scheduled-task orchestration: Cron / fixed-interval / event / queue triggers, APScheduler-driven.
Whole-site crawl → Markdown knowledge base (httpx + html2text), feeding downstream retrieval and RAG.
Generate the four-piece skill scaffold to quickly crystallize new Skill / MCP Server / Coded Agent.
From Agent intent to result return, the RPA Skills System chains defense, extraction, and orchestration into a reusable pipeline.
Upgrade "automation" from scattered scripts into an orchestratable, auditable, reusable skills platform.
The difficulty is not single-point scripts, but engineering fragmented engines and defenses into a platform Agents can stably schedule.
Business logic written once; Skill CLI and MCP process are thin shells, eliminating dual maintenance and behavior drift, minimal cost to add skills.
Converge CloakBrowser / Obscura / BrowserCluster / AgentGo / Lamda into one interface; Agents ignore the underlying differences and compose freely.
Anti-bot, JS reverse, session/fingerprint, credential audit are first-class citizens of the skill matrix, ready out of the box.
workflow-orchestrator + skill-scaffold-generator crystallize repeated flows into reusable, parallelizable, generatable skill assets.
Convert technical capability into quantifiable, reusable business value.
Product content has been published based on internal materials. The following areas are planned for further development:
Explore Xianma AI solutions in other domains