2026-02-24

skill-vs-agents-vs-claude-md - SKILL.md vs AGENTS.md vs .claud Files

Comprehensive comparison of three agent instruction formats: AGENTS.md (universal project instructions for any agent, agents.md spec), SKILL.md (Claude Code workflow knowledge), and .claud files (packaged Claude skills). Quick answer: AGENTS.md is project-level guidance, SKILL.md is workflow documentation, .claud files are individual skill packages.

anthropic-distillation-attacks - AI Model Theft at Industrial Scale

Anthropic identified three Chinese labs (DeepSeek, Moonshot, MiniMax) conducting massive distillation attacks — 16M exchanges via 24K fraudulent accounts to steal Claude's capabilities. Stripped-out safeguards create national security risks (bioweapons, cyber ops, surveillance). Attacks undermine export controls by bypassing chip restrictions through theft rather than innovation.

autonomous-dogfooding-skill - Autonomous Testing for agent-browser

Chris Tate (@ctatedev, Dev @ Vercel) released an autonomous dogfooding skill for agent-browser that explores your app like a user — clicks buttons, fills forms, tests edge cases, checks console, captures repro videos and screenshots, outputs structured report with severity ratings. No test scripts, no manual QA.