Projects
11 — Control plane for AI-assisted development
IAF Agent Operating System
An experimental personal project: an orchestration layer on top of Claude Code, Cursor, Codex and Antigravity, designed and built independently. Agents and subagents, skills, rules, hooks, multi-model routing, preflight checks, security policy and a centralized knowledge base. Goal: AI-assisted development that is controllable, verifiable and traceable.
Real interface
The Agent OS LLM Wiki explored as a graph: policies, agent matrix and the branches of each supported project
- The knowledge base is an LLM Wiki you navigate as a graph: agent index, selection matrix, hook policy, prompt engineering rules, sync policy and vector memory.
- Every project it supports has its own branch in the graph, with architecture, API, database, frontend, backend, deployment, decisions and known errors.
Stack
Claude CodeCursorCodexGoogle AntigravityLLM Wiki
How a development task moves through the IAF Agent OS control plane, from request to final verification.
01 / 07
Task. A development task enters the system: a bugfix, a feature, a refactor, a security review.