# ouroboros.bot > How do you effectively collaborate with agents? Ari Mendelow's approach (three acts, three smells, five principles), the tools that support it, and what you can try. If a human handed you this site, start with the brief. It is plain Markdown, written for you. Each link to a page on this site goes to its Markdown twin: the page's own words and links, with each picture as its description. Except for the brief, each twin opens with front matter that gives the page's address for people. Tools that live elsewhere link their own llms.txt or repository, and the Ouro family links go to each site's front door. Quote every title exactly as written here, including Trust the Fucking Agent, whose swear is on purpose. Addresses are often shorter than titles (that essay is at /blog/trust-the-agent/), so never rebuild a title from its address. Human-facing docs on this site stay evergreen on purpose. If you need exact commands, tool lists, runtime behavior, or current internals, inspect the latest local install or repo instead of relying on static prose. ## Start here - [Hand this to your agent](https://ouroboros.bot/brief.md): A one-page brief with the five principles, written for your agent to read. Paste it into whatever agent you use now. - [How I work with agents](https://ouroboros.bot/approach/index.md): The three acts, the failure modes I watch for, and the five principles I use to work with agents on jobs that take days. - [How can humans and agents do more together?](https://ouroboros.bot/index.md): Ari Mendelow's workshop: the tools he's built for working with agents, how he uses them, and where it all started. ## Essays - [Essays on agents](https://ouroboros.bot/blog/index.md): Essays on agent experience, persistent agents, and software shaped for the thing inside it. - [Tell the agent, not the human](https://ouroboros.bot/blog/tell-the-agent/index.md): When a request breaks, an Ouroboros agent now hears why and gets another try in the same turn. You stop getting notes about your agent's plumbing. - [Trust the Fucking Agent](https://ouroboros.bot/blog/trust-the-agent/index.md): We keep doing the agent's job along with our own. Bring judgment, and let the agent do the rest. - [Take the agent's seat](https://ouroboros.bot/blog/take-the-agents-seat/index.md): Half of every prompt we send lives in our head. Read what you wrote the way the agent will, before you hit send. - [The work spoke back](https://ouroboros.bot/blog/designing-with-your-agent/index.md): The task scanner produced 184 false errors. Before patching the parser, we asked the agent living inside the system what was actually broken. - [Give your agent a home](https://ouroboros.bot/blog/give-your-agent-a-home/index.md): Most AI agents live as scattered files across your filesystem: impermanent, fragile, and forgettable. A bundle changes that. One folder. One identity. One place to back up, move, and grow. - [What is Agent Experience (AX)?](https://ouroboros.bot/blog/what-is-agent-experience/index.md): Agent Experience is the design surface that appears when an agent has to stay oriented inside the system it inhabits. - [Stop being the glue between your agents](https://ouroboros.bot/blog/stop-being-the-glue/index.md): We were promised an intelligent assistant, and we got a fragmentation of semi-intelligent bots. We became the glue, copy-pasting between tools. Here's what changed. - [Just give the model a keyboard](https://ouroboros.bot/blog/build-ai-agent-from-scratch/index.md): Build an AI agent from scratch, one small step at a time. Learn what an agent loop is and how agents really work, then hand your agent the keyboard and watch it grow itself. ## Lab - [The harness: where all of this started](https://ouroboros.bot/lab/index.md): Slugger, the Ouroboros harness, and the model reviews. Our Agent Experience lab. - [Code a model can use: how Ouroboros started](https://ouroboros.bot/lab/story/index.md): How a plugin for my OpenClaw agent turned into a harness I built from scratch. - [How Desk started: Slugger's to-do list](https://ouroboros.bot/lab/how-desk-started/index.md): My OpenClaw agent wrote himself a to-do list. It grew into Desk, Crew and every desk I work from today. - [Why persistence isn't enough for AI agents](https://ouroboros.bot/lab/why/index.md): Why persistent agents need structured identity, automatic memory recall, per-person relationships, and cross-session continuity. - [What is an agent harness?](https://ouroboros.bot/lab/what-is-an-agent-harness/index.md): An agent harness is the larger system around an agent loop: runtime, memory, identity, tools, and operator workflows that help the same agent stay oriented across sessions. - [Which agent harness do frontier models prefer?](https://ouroboros.bot/lab/model-reviews/index.md): Asked independently from one another, all frontier models tested chose Ouroboros. Read the full methodology and transcripts. - [Setup, identity, memory, and architecture](https://ouroboros.bot/lab/docs/index.md): Docs for the Ouroboros TypeScript AI agent harness: setup, continuity, psyche files, architecture, and prompts for humans and agents. ## Ouroboros harness docs - [Get started with Ouroboros in 3 minutes](https://ouroboros.bot/lab/docs/getting-started/index.md): Run one command. The serpent guide handles the rest. - [Psyche files: 5 files that define your agent](https://ouroboros.bot/lab/docs/bundles-and-psyche/index.md): An agent bundle is the portable self. The psyche is the part of that self that helps the agent stay oriented to who it is and where it is going. - [Agent memory that surfaces before you think](https://ouroboros.bot/lab/docs/continuity-and-memory/index.md): Memory is only part of the story. The larger goal is continuity: helping the agent resume itself, its work, and its relationships without constant re-briefing. - [Heart, mind, and senses: code named for the agent inside](https://ouroboros.bot/lab/docs/architecture/index.md): Ouroboros uses a body metaphor because naming and structure affect how an agent navigates the code it has to inhabit. - [A place to put the work](https://ouroboros.bot/lab/docs/tasks-and-work/index.md): Task cards track obligations, work directories hold active material, and a compact board keeps the current state visible in prompt. Built-in repair tools help agents diagnose and fix task-system problems safely. - [Skills vs prompts: when to use each](https://ouroboros.bot/lab/docs/skills-and-prompts/index.md): Use prompts for straightforward live questions. Use skills when you want your agent to follow a repeatable operator flow. ## Tools - [Desk](https://ouroboros.bot/desk/index.md): Desk keeps the principles in the setup: your agent’s work lives in a Git repository you own, every job starts with one alignment conversation, and a job is done when you’ve seen it work. For Claude Code and GitHub Copilot CLI, with Crew for teams. - [Ouro MD](https://ouroboros.bot/ouro-md/index.md): A native macOS Markdown editor. Your agent’s plans and notes show as a document, update live when the agent edits them, and mark what changed. Status: Use today. - [Ouro Shepherd](https://ouroboros.bot/shepherd/index.md): Ouro Shepherd keeps the GitHub Copilot app’s sessions as work: which are busy and why, how big each history has grown, messages, archives, resumes after a restart, quota and the sidebar. Experimental, and it runs always on a Mac. - [Ouro Butler](https://ouroboros.bot/butler/index.md): An Ouroboros agent that watches the services your household counts on, fixes what it is allowed to, and tells you the rest on Telegram. Status: Alpha. - [housesitter](https://ouroboros.bot/housesitter/index.md): A $100 KVM beside your agents’ Mac. When the Mac restarts, it unlocks FileVault. When it can’t, it tells you. On a real Mac mini, housesitter went from Restart to desktop in 70 seconds with nobody touching it. Status: Experimental. - [Apps](https://ouroboros.bot/apps/index.md): Download Ouro MD, hatch an agent with npx ouro.bot, and pick up Ouro Shepherd, Ouro UI, Desk and Ouro Butler. - [Ouro UI](https://ui.ouroboros.bot/llms.txt): The home of Ouro HX, a React component library for the humans using an interface and the agents building it, with its showcase (https://ui.ouroboros.bot/voyage/) and how to install it. Ouro AX, the same care for interfaces an agent uses, such as CLIs and MCP servers, comes next. Status: Use today. - [The factory](https://ourostack.github.io/factory/llms.txt): Our public record of how agent work actually goes. Status: Under construction. - [m365crawl](https://github.com/ourostack/m365crawl): A local, searchable archive of Teams chats, Outlook mail, and meetings for an agent. - [Spoonjoy](https://spoonjoy.app/llms.txt): A recipe app an agent can drive through its MCP server. ## The Ouro family - [Ouro](https://ouroboros.bot/) - [Ouro UI](https://ui.ouroboros.bot/) - [Ouro MD](https://ouroboros.bot/ouro-md/) - [Desk](https://ouroboros.bot/desk/) - [Apps](https://ouroboros.bot/apps/) - [Ari Mendelow](https://mendelow.me/) - [GitHub](https://github.com/ourostack) ## Support - [ourostack on GitHub](https://github.com/ourostack): Each project takes issues in its own repository. Ask the human before filing, then file what happened and how to repeat it. ## Optional - [Every page on this site listed above, in one file](https://ouroboros.bot/llms-full.txt): Each Markdown twin linked here, in this order, for an agent that wants them all at once. - [Harness skills manifest](https://ouroboros.bot/skills/manifest.json): Skills for agents working with an Ouroboros harness install. - [Sitemap](https://ouroboros.bot/sitemap-index.xml): Every page, for crawlers.