Hunting bugs with a team of agents
I pointed a team of Claude Code agents at one of my codebases and asked them to find bugs. Here is the prompt I used, how I tightened it, and what came back.
Avasdream Blog & Streams
Minimze Platform Risk by Building in Public and on multiple platforms.
Latest posts
Wrapping Claude CLI for Agentic Applications
How to build production-ready wrappers around Claude CLI. Input/output formats, structured JSON schemas, permission modes, and everything you need for CI/CD and agent pipelines.
Using Claude Code and Agent SDK with Alternative Providers
Configure Claude Code and the Anthropic Agent SDK to work with Z.AI, OpenRouter, or other providers instead of the default Anthropic API.
Automating Development with GitHub Agent Loops
Build an autonomous development loop that picks up GitHub issues, implements them with TDD, runs QA, and closes issues automatically using Claude Code Agent Teams.
Automating Development with GitHub Agent Loops
Build an autonomous development loop that picks up GitHub issues, implements them with TDD, runs QA, and closes issues automatically using Claude Code Agent Teams.
What Gives AI Text Away (And How to Fix It)
AI-generated text has tells. Wikipedia editors catalogued 28 of them. Here is what they found, why it happens, and how to make AI writing sound human again.
The Implementation Intent Planner: A Skill for AI-Assisted Development Planning
A structured skill that transforms how AI agents create implementation plans. Learn why each component exists and how it prevents common planning failures.
Inside OpenClaw: How the Agent System Routes Models, Tools, and Skills
A technical deep dive into OpenClaw agent internals — 15+ AI providers with auto-discovery, a 9-layer tool policy system, context window management with cache-aware pruning, the skills loading pipeline, and Docker sandboxing with zero-trust defaults.
Inside OpenClaw: Browser Automation, Media Processing, and the Canvas System
A technical deep dive into how OpenClaw controls browsers via Playwright and CDP, processes media through a multi-provider AI pipeline, renders interactive canvases on mobile nodes, and synthesizes speech with three TTS backends.
Inside OpenClaw: The Channel & Messaging System That Speaks Seven Protocols
A technical deep dive into OpenClaw channel internals — the two-tier dock/plugin abstraction, plugin registry architecture, inbound normalization to MsgContext, outbound adapters with lazy loading, and platform-specific handling across Telegram, Discord, WhatsApp, Signal, Slack, iMessage, and LINE.
Inside OpenClaw: CLI Boot Sequence, Command Architecture, and the Terminal UI
A technical deep dive into OpenClaw CLI internals — fast-path routing that bypasses Commander entirely, lazy subcommand loading via placeholder-and-swap, the full command registry, profile isolation, the component-based TUI with WebSocket streaming, the onboarding wizard with its prompter abstraction, and plugin CLI extensibility.
Inside OpenClaw: Core Architecture & the Gateway Control Plane
A technical deep dive into OpenClaw core architecture — the Gateway daemon as a WebSocket control plane, the bootstrap sequence, protocol framing with AJV-validated schemas, JSON5 configuration with hot-reload, connection management with backpressure, and the single-user local-first design philosophy.
Inside OpenClaw: Infrastructure, Security Subsystems, and the Plugin SDK
A deep dive into OpenClaw infrastructure — cron scheduling with stuck job detection, 7 security subsystems including Ed25519 device identity and SSRF prevention, the plugin SDK with 15 lifecycle hooks, node pairing with crypto verification, and daemon management across platforms.
Inside OpenClaw: Sessions, Sub-Agents, and Multi-Agent Orchestration
A technical deep dive into OpenClaw session management — hierarchical key resolution, file-based persistence with JSONL transcripts, sub-agent spawning with lifecycle tracking, cross-session communication via A2A ping-pong, and the security model that keeps it all isolated.
How OpenClaw Remembers: A Deep Dive into AI Agent Memory Architecture
An end-user guide and technical deep dive into OpenClaw memory system — how plain Markdown files, hybrid vector search, atomic reindexing, and pre-compaction flushes give an AI assistant persistent, searchable memory across sessions.
Hardening a VPS for OpenClaw: SSH, fail2ban, and Security Validation
A practical guide to securing a fresh VPS, installing OpenClaw, and validating your setup. Covers SSH hardening, fail2ban with real attack data, initial OpenClaw configuration, and prompt injection awareness.
Validating OpenClaw Security with Shodan
Using Shodan to find OpenClaw instances and verify they are properly secured. Spoiler: they are.
Prompt Injection Defenses: What the Research Actually Shows
An evidence-based analysis of prompt injection defenses based on 150+ academic papers. Single defenses achieve 45-60% effectiveness; multi-layered approaches reach 87-94%. The attack surface in AI agents like OpenClaw is massive—and the consequences are devastating.
Custom Status Line for Claude Code: Track Context and Costs in Real-Time
Set up a custom status line in Claude Code that displays context window usage, session costs, model name, and git branch - all color-coded for quick visibility.
Test-Driven and Spec-Driven Development: Building Lossless Feedback Loops for Agentic Coding
An exploration of how Test-Driven Development and Spec-Driven Development create lossless feedback loops in AI-assisted coding, with practical examples using Claude Code.
OpenSpec vs Spec Kit: Choosing a Spec-Driven Development Framework
A comprehensive comparison of OpenSpec and GitHub Spec Kit. Two open-source frameworks for structured AI-assisted software development.
Claude Skills: A Technical Deep Dive into Agentic Orchestration
An in-depth technical analysis of Claude Skills architecture, progressive disclosure mechanism, and how Skills complement the Model Context Protocol for building autonomous AI agents.
Delete All Vercel Projects with a Node.js Script
A Node.js script to bulk delete all Vercel projects using the REST API, with pagination handling and rate limit management.
Building a Documentation-to-Skill Generator for Claude Code
Create a metaskill that automatically converts documentation files into structured Claude Code skills, enabling instant access to any framework or library documentation.
Intercepting Claude Code Traffic with Burp Suite
A step-by-step guide on how to configure Claude Code to route traffic through Burp Suite for inspection and security research.
Essential Resources for HackTheBox and Kali Linux Setup
A curated list of essential links and resources for setting up Kali Linux with VirtualBox and getting started with HackTheBox.
CSS Minecraft: The Most Insane CSS-Only Game Ever Created
How Benjamin Aster built a fully playable Minecraft clone using zero JavaScript - just pure HTML and CSS wizardry.
Building a Git Contribution Bot with GitHub Actions: Automating Your Green Squares
Learn how to create an automated Git contribution bot using GitHub Actions that randomly commits to maintain your GitHub activity chart with green squares.
Complete Guide to Setting Up DVWA and Mastering Web Vulnerabilities
A comprehensive beginner-friendly guide to setting up DVWA with Docker and mastering web vulnerabilities from SQL injection to command injection across all difficulty levels.
Fixing Vercel "No Serverless Pages Were Built" Out of Memory Error
How to resolve the "No serverless pages were built" error caused by out-of-memory issues during Vercel deployments.
Fixing "Cannot Read Properties of Undefined" in Expo iOS Submission
A quick fix for resolving the "Cannot read properties of undefined" error when submitting an iOS app using Expo EAS CLI.
FastAPI to Google Cloud Run: Prepare, Deploy, and Automate
Learn how to prepare, deploy, and automate the deployment of a FastAPI application to Google Cloud Run.
Data Analysis Wahl-O-mat: Part 1
Analyze political party positions using data from the Wahl-O-Mat tool for the upcoming European elections.
Transcribe Audio Input with Firebase Firestore and Storage through OpenAI
Learn how to transcribe audio input with Firebase Firestore and Storage through
Implementing Feature Flags for Subscriptions in Next.js
Learn how to implement feature flags based on subscription plans in a Next.js application using a simple JSON data structure.
Unlimited and shared command line history on MacOS
How to set up unlimited and shared command line history on MacOS using zsh.
Solving Firebase Emulator Error: Failed to Load Function Definition
Learn how to resolve the "Failed to load function definition from source" error when using Firebase Emulators.
Merging Multiple Git Repositories into a Monorepo Without Losing History
A step-by-step guide on how to merge multiple Git repositories into a single monorepo without losing their commit histories.
React Cheatsheet
React Cheatsheet. A collection of useful React snippets.
DevOps/Kubernetes Cheatsheet
My personal cheatsheet for various DevOps related tools. If there is some context missing here or there, dont hesitate to reach out to me on Twitter.
Automate your GitHub Contributions with GitHub Actions
Automate your GitHub Contributions with GitHub Actions.
Longhorn Namespace Stuck Terminating - Delete Longhorn from Kubernetes Cluster
How to delete Longhorn from a Kubernetes Cluster when the namespace is stuck in "Terminating".
Gatsby Hydration Issue
Fixing the hydration issue in Gatsby. The CSS is not shown correctly in production mode.
Downgrade NodeJS with Brew on Mac OS
How to downgrade NodeJS with Brew on Mac OS.
A Scientific Approach to Context Engineering
An evidence-based analysis of LLM context utilization based on 11 academic studies. Learn why your 200K context window might only be 50% effective, and how to engineer prompts that actually work.
Multi-Cloud Orchestration with Python
I used the strategy pattern to orchestrate operations on worker instances in multiple clouds while exposing only one interface to clients. Code can be found [here](
OSCP: Try harder and try smarter!
A collection of resources and tips to pass the OSCP exam.
Deploy a Jitsi Meet Server on AWS with Terraform
How to deploy a Jitsi Meet Server to AWS with Terraform.
Pentesting: Local file inclusion to remote code execution on Hackazon
Escalating a local file inclusion vulnerability to remote code execution on the Hackazon application. Hackazon is a vulnerable application and the source code is public at GitHub.
Setting up a web application penetration testing laboratory with Terraform and Docker
How to build a beginner-friendly web application penetration testing laboratory in the Amazon cloud using Terraform and Docker.