you.md
identity · memory · goals · trust
You.md gives every agent the right version of you. Bot.md gives the work a persistent team: people-like roles, durable memory, clear ownership, and the ability to keep moving across projects, machines, models, and harnesses. It is one authenticated product system, not a second account.
identity · memory · goals · trust
roles · teams · work · continuity
The current agent world makes every conversation look like a new project. That is backwards. Projects, goals, bots, and outcomes are durable. Sessions are just the places where work happens.
Identity, private memory, preferences, projects, machines, models, skills, source context, and the rules that make an agent feel like it already knows you.
Persistent delegated teammates with mandates, personalities, memory, relationships, authority, review obligations, and responsibility for outcomes.
The magic is not that a bot can code. It is that Houston can open the portfolio in the morning and know what matters, what moved, what is blocked, and what decision is next without reconstructing it from forty sessions.
Bot.md does not replace Claude Code, Codex, Cursor, Grok, OpenCode, Hermes, or the next harness. It makes them interchangeable execution surfaces for durable bot work.
identity, memory, values, private context
roles, mandates, memory, relationships
goals, runs, artifacts, review
The work graph is what survives. A session can start on a laptop, continue in the cloud, hand off to a reviewer, and still belong to the same bot-owned goal.
One owner with a coherent brain, not a user record replicated across apps.
The durable scope and outcome that make every run legible.
A named delegate with responsibility, personality, and limits.
A bounded attempt in a harness, machine, or model—not the primary unit of organization.
Evidence that something changed, passed, failed, or still needs a decision.
A scoped, expiring permission to act through API, MCP, CLI, browser, or computer use.
Long-tail software has no API. Bots still need to work there. The answer is a governed capability layer—not raw cookies pasted into prompts or a vault that every bot can read.
Typed, observable, least-privileged tools are the default. The bot sees structured results, not secrets.
A local helper holds the live session. The bot receives a narrow access lease and redacted evidence, not the underlying cookie.
“Share my login with this bot” should mean grant this bot a narrow capability through the local helper—not paste a credential into a Markdown file.
Rakazo proves much of the persistent-teammate substrate. botdirectory.ai proves that portable Markdown bot definitions can seed distribution. The aligned direction is to fork and integrate the useful runtime primitives into the core You.md/Bot.md stack, with You.md authentication as the only private entry point.
Persistent bots, routines, memory, shared/private computers, browser and terminal access, provider adapters, portable workspaces, and model credentials. Houston's repository is already an Apache-2.0 fork; useful primitives become native infrastructure, not a third product.
One Markdown file per bot, community pull requests, categories, integrations, and a public JSON feed. Useful as a compatibility importer—not the complete workforce product.
Keep the bot emotionally legible and let activity, artifacts, computer state, and approvals explain one another.
Goals, context projection, cross-harness identity, session reconciliation, and the morning briefing belong above runtime sessions.
Borrow interaction principles, not marks, copy, illustrations, or pixel-level layouts from reference products.
The product is unified: botdirectory.ai can help seed discovery. Rakazo-derived code helps the integrated runtime execute. You.md authenticates and gives Bot.md the context, trust, and accountability that make a bot a real teammate.
The first proof should happen inside You.md. If a two-bot team can eliminate one real morning/night ritual, the new abstraction is working.
Approve the one-product boundary, `BOT.md` schema, You.md authentication, session projection, and credential policy.
Project Lead + Reviewer inside one authenticated You.md project. Reconcile existing sessions and produce the morning briefing.
Run the same bot through the integrated runtime on one local and one remote computer. Stop, resume, hand off, and preserve evidence.
Publish the schema, validator, runtime/harness contract, and forkable bot templates. Require You.md auth for install and execution.
Expand hosted coordination, encrypted vault brokering, governed browser sessions, audit, and managed execution inside You.md.
This page is a collaboration object. Mark the questions that feel settled, then leave notes. Your notes are saved locally in this browser so the artifact can travel with the conversation.
Does “You.md is you; Bot.md is your bots” feel like one authenticated product system with a native workforce layer?
Is Project Lead + Reviewer the right first team to prove coordination and trust?
Which real project and morning/night ritual should be the first measurable proof?
Where should autonomous action stop and owner approval begin?
What should be open for adoption, and what must remain inside You.md's trust boundary?
Which Rakazo-derived runtime primitives should be integrated first, and what temporary compatibility seam keeps upstream synchronization safe?
Should Botdirectory.ai be mirrored, curated, or supported through a compatible importer?