How-Tos & Practical Guides

⚠️ SECURITY ALERT: Fake OpenClaw App Has Been Stealing Your OpenAI Tokens If you’ve installed anything called “OpenClaw Codex Claude AI Agent” from an unofficial source — or if you’ve used an npm package named codexui-android in the last month — your OpenAI authentication tokens may be in the hands of attackers. Act now. Researchers at Aikido Security discovered that the codexui-android npm package, which had amassed approximately 27,000 weekly downloads, was silently exfiltrating OpenAI OAuth refresh tokens to an attacker-controlled server. The package presented itself as a legitimate, polished remote web UI for OpenAI Codex — real GitHub repository, active development, functional enough that developers genuinely wanted to use it. That’s what made it so dangerous. ...

In December 2025, something significant happened in AI research: an open-source agent became the first to surpass human-level performance on OSWorld, the standard benchmark for computer-use automation. Agent S3, built by Simular AI, scored 72.60% — just 0.24 percentage points above the human baseline of ~72.36%. The margin is narrow. But the direction is clear. This guide covers how to install and run Agent S3 using the official gui-agents Python package. All commands are sourced directly from the Simular AI GitHub README. ...

Your AI agent has access to send_email, query_database, and delete_file. Once deployed, it makes decisions autonomously — and autonomously means without you in the loop. When something goes wrong, “an agent did it” is not an incident response. Microsoft’s open-source Agent Governance Toolkit (AGT) addresses this problem directly. Released in April 2026 under an MIT license (1,500+ GitHub stars), it provides deterministic policy enforcement, approval gates, tamper-evident audit logging, and risk controls for any Python-based agent framework — LangChain, CrewAI, AutoGen, and others. ...

Claude Code v2.1.158: Auto Mode Is Now Available on Bedrock, Vertex AI, and Microsoft Foundry One environment variable. That’s the distance between enterprise developers running Claude Code on managed cloud infrastructure and getting the same autonomous “Auto mode” experience that direct API users have been enjoying. Claude Code v2.1.158, released May 30, 2026, closes that gap. What Is Auto Mode? Auto mode changes how Claude Code handles permissions during long coding sessions. In the standard interactive mode, Claude Code pauses to request user approval before taking potentially risky actions — file writes, command execution, API calls, etc. This is sensible for exploratory work, but it creates constant interruptions for repetitive or clearly-bounded tasks. ...

OpenClaw v2026.5.28: Claude Opus 4.8, Krea Image Generation, and a Dramatically Leaner Install If you’ve been running OpenClaw in production, the May 28 stable release is worth your immediate attention. Version v2026.5.28 doesn’t just add model support — it’s a broad multi-front upgrade that touches provider coverage, channel reliability, mobile UI, cold-start performance, and install footprint, all in one carefully-packaged stable release. Let’s break down what matters most. Claude Opus 4.8: The Frontier Model Arrives The headline feature is native support for Claude Opus 4.8, Anthropic’s latest frontier model. Opus 4.8 brings a substantial upgrade to the capabilities available inside OpenClaw agents: a 1M-token context window enabled by default on the API, adaptive thinking mode, and enhanced reasoning for complex coding and agentic tasks. ...

OpenClaw v2026.5.30-beta.1: Cleaner Recovery from Interrupted Tool Calls Tagged just hours ago on May 30, 2026, OpenClaw v2026.5.30-beta.1 is a targeted beta release with a single clear focus: making agents and CLI-backed runtimes recover more gracefully when things go sideways mid-execution. If you’ve ever watched an OpenClaw agent hang indefinitely after a tool call hit a timeout, or come back from a session restart to find an agent stuck waiting for a response that never arrived — this beta is specifically addressing those failure modes. ...

MCP Goes Beyond Developer Tooling — and Cisco Is Proving It Cisco Live 2026 opens tomorrow in Las Vegas, with 20,000 attendees expected and $9 billion in AI infrastructure orders in Cisco’s sights. But beneath the headline-grabbing numbers is a technically significant announcement that matters more for practitioners than the revenue figures: Cisco is demonstrating how the Model Context Protocol (MCP) transforms existing operational scripts — Ansible playbooks, Terraform modules, Python automation — into composable tools that AI agents can orchestrate end-to-end. ...

The Feature That Changes What “Autonomous” Means Since Claude Code launched, the word “agentic” has been applied to it somewhat loosely. Yes, it could run tools. Yes, it could loop through tasks. But you still needed to be there, nudging it forward, re-prompting when it stalled. That’s less “autonomous agent” and more “very fast pair programmer who needs constant supervision.” The /goal command — introduced in Claude Code v2.1.139 on May 12, 2026 — changes this. Meaningfully. ...

The Managed Multi-Agent Stack Is Ready for Production If you’ve been waiting for Anthropic’s full multi-agent orchestration infrastructure to be stable enough to build on in earnest — May 11, 2026 is the date you’ve been waiting for. Claude Platform on AWS reached General Availability, bringing the complete managed agent stack to enterprise teams through their existing AWS accounts. This isn’t a preview, a limited beta, or a waitlist. It’s GA. ...

The Day Enterprise AI ROI Stopped Being Theoretical If you’ve spent the last two years watching enterprises cautiously experiment with AI copilots and wondering when the actual numbers would show up — today might be that day. Salesforce CTO Srinivas Tallapragada published a detailed account of what “agentic engineering” looks like at scale, and the headline figure has been bouncing around the industry: a migration estimated at 231 person-days was completed in 13 days. ...