Compound Agent OS
One person, multiplied by a system that learns.
Compound Agent OS is my personal AI operating system: a growing roster of specialized agents, shared skills, and persistent memory that lets one person operate with a team’s worth of leverage. It enables both compounded productivity and compounded improvement. Each refinement to an agent, skill, workflow, or memory layer makes the whole system more capable over time.
Compound Console
The agents below don’t just live in markdown files — they run inside Claude Code, Cursor, Codex, and Compound Console, a custom orchestration platform I built on the Claude Agent SDK. Type a goal, and the orchestrator reads the roster, assembles a plan, and dispatches the right specialists in the right order. Every agent streams its output in real time so you can see exactly what’s being researched, written, or built.
Under the hood: agents are authored as plain markdown with YAML frontmatter (no vendor lock-in), prompt caching cuts API costs by ~90% on repeated runs, composable skill modules attach to agents by task-relevance scoring, and each session writes learnings back to project context so the next run starts smarter. General mode runs agents in parallel toward a synthesized deliverable; pipeline mode chains them through a gated architect → engineer → code reviewer sequence.
The Roster
Each agent is purpose-built for a specific domain — and designed to work alongside the others.