14 min read—
NVIDIA SkillSpector: Scan AI Skills Before Installing
NVIDIA SkillSpector scans Claude Code, Codex and MCP skills before install, scores risk 0-100 and flags SAFE, CAUTION or DO NOT INSTALL, with a CI guide.
Posts
All the articles I've posted.
- 10 min read—
Claude Code Remote on AWS Bedrock AgentCore
Run Claude Code remotely on AWS Bedrock AgentCore: one microVM per task, model calls only via Bedrock, and a pull request when done. Setup guide and limits.
- 9 min read—
Hex Local Voice Dictation: Speak and It Types, No Cloud
Hex is a free, open-source Rust dictation app from Anomaly: hold a key, speak, release, and text pastes into the focused app. Guide for macOS, Windows, Linux.
- 12 min read—
Cut LLM Costs 4x with a Jev → Opus Cascade
Jev decides first and escalates only uncertain cases to Claude Opus 5.5: 4.39x cheaper on Hightouch's easy test, 3.5x on Banking77, accuracy almost unchanged.
- 13 min read—
GPT-6 Cyber Found 948 Linux Kernel Privilege Bugs
GPT-6 Cyber found 948 Linux kernel privilege-escalation flaws with candidate patches. What the number really means, plus the patch checklist for sysadmins.
- 13 min read—
Anthropic on Agent Memory: Notes, Wikis, and Dreaming
Anthropic's agent memory architecture: transcripts, notes, wikis, and 'dreaming' that fixes wrong notes. CL-Bench, Pokémon, and wiki results, plus six rules.
- 10 min read—
Taste Skill: Stop AI Slop in AI-Coded Websites
Taste Skill 2.5.1 helps Claude Code and Codex build websites without AI slop — vetted palettes, layouts, real photos, and a 5-viewport browser checker.
- 9 min read—
Do AI Agents Really Work? The Real Numbers
DoorDash got a few hundred agent orders, Uber just four rides, Airbnb almost no agent bookings at all. What do the real numbers say about the AI agent hype?