Applied AI engineer · Atlanta, GA
I build production AI agents, modular Claude Skills, and the reproducible runtime environments that let teams ship them. My work today is in production at CloudWarriors, on my own SaaS, and across nine teammates' chat surfaces. This site is where the work lives in the open.
Every morning a Claude Skill I built (ecosystem-update) crawls GitHub, arXiv, and the Claude community for new agent patterns, hooks, skills, and research. It diffs each item against my current setup, implements safe Quick Wins automatically, and writes a dated digest. I share it with my team. Now I'm sharing it with you.
Each digest ships in two views: a public digest (curated TLDR with sources) and a personal digest (my own adoption decisions and what I rejected with reasons). Pick whichever matches how deep you want to go.
My live ~/.claude runtime — agents, skills, hooks, route classifier, planning gates, postflight checks, governance config. The same harness my CloudWarriors teammates run.
My live ~/.Codex runtime — Codex skills, hooks, agents, control-plane scripts, dual-brain dispatch wired to OpenAI Codex. Companion to chad_claude_configs.
Open-source audio recording + transcription application. Hosted demo at clio.chadacus.dev.
The full catalog — claude-squad and codex-squad (legacy predecessors), claurst (Rust), project-management-app, automation-station, and more.
Autonomous engineering captains, persona agents on chat surfaces, marketing automation that drafts in my voice. Built with saturation gates, circuit breakers, and human steering channels — they run unattended, they don't run away.
Forty Skills in daily production, each a self-contained markdown primitive with documented inputs and outputs. Composable: outputs of one Skill become inputs to the next.
A hardened Claude + Codex multi-agent runtime — route classifier, governed dispatch budgets, planning gates, postflight checks, dual-brain dispatch (planner-reviewer + cheaper implementer), rubric-gated acceptance. My team runs it, kept current by the ecosystem-update Skill.
omni-mem is the memory layer underneath every Claude, Codex, and Ollama session I run. Deterministic retrieval, event-sourced JSONL, MCP server in Docker. Public mirror coming.