OPEN X SECURITY
RESEARCH - NOTES FROM THE REPOS

Research

Product updates and engineering write-ups on managing AI coding agents - configs, deploys, usage limits, and what we learned building it.

FEATUREDAUG 2026 · 9 MIN READ AgentHarbor vs ccusage vs CodexBar: a monitor reads, a control plane writes ccusage and CodexBar are excellent at reading your AI coding usage. AgentHarbor reads it too, then deploys your configs, locks down permissions, wipes a secret you pasted into a chat by mistake, and switches your models. An honest comparison, including where the other two win. READ THE POST ↗ The AgentHarbor tray with live limits and cost for Claude, Cursor, Codex and Gemini
CVE-2026-28931 in the macOS kernel NFS client AUG 2026 A server said zero, and the macOS kernel believed it: CVE-2026-28931 SECURITY RESEARCHAPPLE The AgentHarbor registry, one searchable library of 119 skills, rules, MCP servers, agents, and hooks AUG 2026 Why I built AgentHarbor: a year of fixing one annoyance at a time PRODUCTAGENTHARBOR The AgentHarbor Secrets Manager storing API keys in the Keychain, values hidden until revealed JUN 2026 How AgentHarbor keeps your configs safe: backups, drift detection, and undo ENGINEERINGAGENTHARBOR AgentHarbor agents deployed from one markdown file to Claude Code and Cursor JUN 2026 Introducing AgentHarbor: one harbor for all your AI agents PRODUCTAGENTHARBOR