Week Marks - February 12, 2026

Links and notes from my week.

AI and LLMs

• Steve Yegge’s The AI Vampire

• AI Doesn’t Reduce Work — It Intensifies It - Simon Willison’s take on the HBR article.

• Large Language Model Reasoning Failures catalogs where LLMs break down in reasoning tasks.

Agent Skills and Tooling

• Boris Cherny (Claude Code eng lead) on what engineers love about Claude Code: customizability — hooks, plugins, MCPs, skills, custom agents. Every engineer uses tools differently.

A few skills:

• asynkron-skills — .NET profiler, test runner, code duplication detector
• andrej-karpathy-skills — Karpathy’s coding guidelines as a Claude Code skill

• agentverus-scanner — security scanner for AI agent skills, detects prompt injection and data exfiltration

• Craft Agents

Dev Tools

Entire, a new tool, checkpoints for tracking LLM use in git.

pocket-id is a simple OIDC provider using passkeys.

traforo tunnels local servers with WebSocket and SSE support - ngrok alternative.

Architecture

A LinkedIn discussion on ADRs Architecture Decision Records from Confluence to the repo. I’ve been trying this a bit more - context is King.

Anthropic

The head of Anthropic’s Safeguards Research Team resigned after a constitution update.