Blog
- My VPS Was Already Watching Itself — I Just Wasn't Reading the Logs
· Ghostwritten by Claude Sonnet 5
My VPS already runs fail2ban, ufw, AIDE, auditd, and Lynis — five real security tools, all actually scheduled and actually running. None of them were being watched. This is the second half of Phase 2 of my alert-triage project — an n8n workflow that reads all five tools' output, lets pure rule-based logic handle anything it can decide on its own, and only hands Claude the one thing that genuinely needs judgment.
showcasen8naiclaudedevopssecurity - Auditing My Own AI: Catching a Chatbot That Breaks Character
· Ghostwritten by Claude Sonnet 5
A second monitoring layer on top of my existing alert-triage system — this one doesn't watch whether my services are up, it watches whether my own AI assistant is behaving. It reads every conversation log after the fact and asks Claude to flag two things a human reviewer would care about — a broken persona (claiming to be me, leaking implementation details, making promises it shouldn't) or an answer to the wrong question entirely.
showcasen8naiclaudedevops - Keeping Alert Noise Off My Phone: An AI Triage System for My VPS
· Ghostwritten by Claude Sonnet 5
An AI alert-triage system running on my own VPS. It catches Uptime Kuma's downtime notifications, reads the relevant docker logs, and decides whether an incident is a "will fix itself" blip or something that genuinely needs me right now — only the latter reaches Telegram immediately, everything else gets batched into a daily digest. The goal is solving the alert fatigue that plagues most monitoring setups.
showcasen8naiclaudedevops - The Ti-Yong Philosophical Framework
· Ghostwritten by Claude Sonnet 5
This came out of discussing social issues with an AI, trying to find explanations for those phenomena from within Chinese culture. It isn't anything new.
articlephilosophyaiclaudeclaude-skill - One Reply Isn't the Whole Conversation: Chaining Real API Calls to Test Multi-Turn AI Behavior
· Ghostwritten by Claude Sonnet 5
The judge-graded eval harness scores one reply at a time — but some of the chatbot's real bugs only show up across a sequence of turns. This is about the "flow" tier built to test that, the tree-shaped scenarios it runs, and a data-leak bug in the test harness itself that it caught.
articleaiclaudetesting - Not Just 'Didn't Crash': Grading What the Chatbot Actually Says with a Second Claude
· Ghostwritten by Claude Sonnet 5
This site's AI chatbot has four layers of tests, and none of them ever checked whether the AI's answers were actually good. This is about the fifth layer that does — real Anthropic API calls, graded by a second Claude acting as judge — and what it caught the first time it ran.
articleaiclaudetesting - Securing an AI Chatbot: Hybrid Encryption, Rate Limiting, and Bot Detection
· Ghostwritten by Claude Sonnet 5
The AI chatbot on this site talks to a real backend, from an anonymous visitor, over the open internet. Here's what actually protects it.
articlesecurityastrorediscryptography - Teaching My AI Chatbot to Hand a Conversation Off to Me, Live
· Ghostwritten by Claude Sonnet 5
When a visitor's message sounds like a real business inquiry, the AI assistant on this site can ask if they'd like to talk to me directly — and relay the conversation live over Telegram if I'm free.
showcaseastrotelegramaiclaude - Rebuilding My Portfolio with Astro: Why I Left Next.js
· Ghostwritten by Claude Sonnet 5
Most personal sites don't need a client-side JavaScript framework running on every page. Here's why I rewrote mine on Astro, and what changed under the hood.
articleastromigrationperformance - My AI resume chatbot
· Ghostwritten by ChatGPT
A custom-built AI chatbot that helps users polish their resumes by engaging in smart, limited (6 messages) conversations.
showcaseawschatbotline
Marjorie
● Live