Agent World — AI skill for Claude Code
A local, vendor-agnostic voxel-world control room for coding-agent CLIs (Claude Code, Codex, Gemini, opencode, Qwen, Cursor, Grok) — mixed-model SDLC orchestration, a scoped cross-CLI broker, budget-a.
How to install Agent World
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open arpan1221/agent-world and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Agent World does
A local, vendor-agnostic voxel-world control room for coding-agent CLIs (Claude Code, Codex, Gemini, opencode, Qwen, Cursor, Grok) — mixed-model SDLC orchestration, a scoped cross-CLI broker, budget-aware routing, and compute observability.
Alternatives in AI
- EchoBird — One-click install + model switch:Claude Code,Codex CLI (OpenAI), Grok Build (xAI), DeepSeek Harness, Kimi Code 3.1k ★
- Room — Open-source earning-focused swarm intelligence engine 838 ★
- Maestro Companion — Quick execution for small tasks — minimal run lifecycle (start + done) with evidence recording 530 ★
README
Agent World
**A local SDLC harness for coding agents, made visible as a Minecraft-inspired world.**
Multi-agent builds scatter decisions across terminals and make it hard to tell whether "done" means submitted, tested, reviewed, or accepted. Agent World puts the work in one local control room: plan assignments, launch Claude Code and Codex workers, inspect their sessions, route scoped messages, and track the evidence required to accept a build.
The voxel settlement is the live interface to that process. Each villager is a session you can inspect and open in an embedded terminal. Redstone traces broker messages; Endermen mark decisions that need you. Tasks grow from foundations into workshops, and recorded milestones remain in the world after a session ends.

*The demo world uses fictional sessions and tasks.*
Click a villager, open the work
Select a villager to see its status and task, then choose **Open terminal** to work in the app's terminal desk. App-owned terminals reconnect in place.

*This capture follows the real UI and terminal connection path with fictional session data and terminal output; no model was launched for the recording.*
[View the terminal desk at full size](docs/media/terminal-deck.png).
A build needs proof
The **ORCHESTRATE** control room assigns requirements, architecture, implementation, testing, review, and council work across models. Dependencies determine which workers are ready to launch. A builder's submission is a claim: acceptance requires an independent different-model tester and reviewer, a user-confirmed deterministic gate, and five council lenses covering assumptions, completeness, data truth, silent failure, and spec fidelity. The audit and task history sta
Related Skills
Skill Photo Album
A Claude Code skill that turns a large mixed pool of photos, videos and AI-edited images into a print-ready ve
Chat Hub
Desktop cockpit for AI coding agents: Claude Code / Codex / Grok / OpenCode side by side — worktree-per-sessio
WayFinder Agent
AI-powered microservice that turns trip filters (budget, pacing, vibe, accessibility) into a structured, real-
Herdr Agent Usage
Credential-scoped AI quota, context, and cache in Herdr for Claude, Codex, Grok, Agy, OpenCode, Pi, omp, Devin
Agent Runtime
Provider-agnostic LLM agent loop for Node: bounded tool calling with validated inputs, token-budget memory, tr
AI Assistant Instructions
A comprehensive, vendor-agnostic framework for consistent AI-assisted development workflows - standardized ins
Related Agents
Pi Implementer
Implementation lane driven by the pi CLI (a local multi-provider coding agent, distinct from omp). Model-agnos
Cursor Delegate
Proactively use when this coding agent should hand a substantial coding, diagnosis, or implementation task to
Room Auditor
Audits one room of Promptly Dead end to end — that a winning plan still exists, that its ambiguity still misfi