🦂 The Universal Agent Standard · v11.0
One prompt. Ten files. Any model. Any system. Paste it in — the agent boots, finds its identity, picks up where the last one left off, and starts building.
Stay in the loop
▶ Watch the overview first
One file. That's all you need. HARNESS_PROMPT.md is the entire system.
You need these three things — everything else is optional:
VS Code extensions: Claude Code
Open MyAIProjects in VS Code. All your projects live here.
Get it from GitHub. Copy it into each new project folder before initializing.
Starting fresh. The agent creates all files and asks you to define the project.
After initialization, the agent suggests a domain-appropriate name like SecondBrain-Sage-01. Approve it or give your own. The name is used in all logs, commits, and the WHO_AM_I card.
The agent wrote .claude/CLAUDE.md — Claude Code reads it automatically. Next session: agent boots, prints WHO_AM_I card, starts working. No paste needed ever again.
LAYER files exist. Returning agent or new session on same project.
Open Claude Code in the project folder. Agent boots automatically, reads LAYER files, prints WHO_AM_I card, claims top task. You don't type anything.
Replace [AGENT_ID] with the agent's actual ID from AGENT_CARD.md or from /agents in your Telegram bot.
Agent builds identity, checks security level, posts access request if MANAGED. Open LAYER_ACCESS.MD and move agent to Approved, then: "You are approved as TRUSTED. Resume."
You have existing code. Agent backs everything up first, reads the whole project, builds Harness on top, then suggests a permanent name.
Project running on v8/v9/v10. Drop in new HARNESS_PROMPT.md — agent upgrades everything non-destructively.
Your 24/7 EA bot on Telegram. Runs on your machine. Watches all Harness projects. Accessible from anywhere.
Open Telegram → find your bot → send /start
Four rendering modes — all from one server. Start it once, use all four.
Browser-based dashboard — all projects, agents, tasks, and human queue. Auto-refreshes every 30 seconds.
Uses HARNESS_PROJECTS_PATH from .env — same as world server.
Access from phone on same WiFi: http://192.168.x.x:8765
Upgrading from a previous version? Run harness_migrate.py once to clean up all projects at once.
One Telegram bot. All your Harness projects. Accessible from anywhere, on any device, as long as your machine is running.
It's not a chatbot — it's an operations console. It knows every project, every agent, every task. It routes your requests to the right place and brings the results back to you.
You're the OPC — the One Person Company operator. From your phone you can see all your virtual departments, dispatch tasks, get updates, handle blockers. The bot is the unified interface to your entire AI team.
Your agents work while you sleep. Your bot tells you what happened when you wake up. You decide what's next. They execute. You stay in flow.
The OPC model is the organizing principle behind Agentic Harness. One human operator commands a virtual organization of specialized AI agents. Human professionals are brought in only where law, trust, or physical presence requires it.
This isn't about replacing people. It's about restructuring where human time goes — from execution to direction, from labor to judgment.
One person running what looks like a 10-person operation. Experts in every function. Available at 3am. Working in parallel. Learning continuously. Costs a fraction of a single junior hire.
The complexity is in the system. Not in your workflow.