Open-source AI operating practice

AI Usage

AI Usage is a practical operating layer for working with multiple Codex agents at once: branches, worktrees, file ownership, status notes, ports, and stop conditions made explicit before the work spreads out.

The first public artifact is the Parallel Agent Orchestrator skill, a small open-source Codex skill that turns parallel work from improvisation into a reviewable coordination contract.

What this project is

The thought is simple: AI work gets safer when agents know the shape of the shared room. A registry, task briefs, status files, and port claims prevent overlapping edits, mystery servers, and final reports that cannot be merged.

The skill is designed for repository-local coordination. It helps an orchestrator split work into branches or worktrees, give each agent a narrow brief, reserve local services deliberately, and collect useful status before anything is merged or deployed.

Public boundary

This is an operating skill, not a hosted AI account system. It does not create Ronova accounts, read private Ronova ID data, or add a second identity layer.

The skill explicitly keeps merge, deploy, form submission, branch deletion, and spending decisions under human or orchestrator approval.

Its helper script is local and dependency-free. The durable source of truth remains the repository where the skill is installed.

Operating loop

The skill makes parallel AI work explicit before it becomes expensive to untangle.

01

Plan the shared state

Create a repo-local orchestration registry, task graph, branch list, merge order, file boundaries, and stop conditions before spawning agents.

02

Assign narrow work

Give each agent one branch or worktree, allowed files, forbidden files, safe commands, expected checks, and a status file to update.

03

Reserve local services

Claim ports before starting servers, record the owner and command, and avoid testing against whichever localhost process happens to answer.

04

Close with evidence

Stop after one reviewable patch, write the final status, list checks run, and surface conflicts before merge or deployment decisions.

Public source

The public repository contains the Codex skill folder, helper script, reference templates, and installation notes for repo-scoped use.

cp -R parallel-agent-orchestrator .agents/skills/Open GitHub repositoryExternal