dotHuman
Protocol Handbook v0.3

Pain Points We Solve

Agent frustrations almost always trace back to one root cause: the model has no durable memory of your project, so nothing you teach it sticks. Here's what that costs — and what a .human/ workspace gives you instead.

The problems .human solves. Without .human: fresh sessions start blind, lessons vanish after each run, no record of how it was done, locked into one agent's memory. With .human: pick up where you left off, lessons compound into skills, an auditable and shareable trail, portable across any agent.
Figure 1. What .human clears up — the recurring pain of agent work, and what you get instead.

What changes

dotHuman doesn't make the agent smarter — it gives the work a place to live, so progress accumulates instead of resetting. Each row pairs a familiar pain with what changes once there's a .human/ the agent can inherit.

The pain What you get instead
Every session starts from zero
A fresh agent thread has no context from the last one — you re-explain the architecture, conventions, and past choices every time.		 It picks up where you left off
The agent primes itself from the .human/ workspace, inheriting your conventions instead of re-briefing from scratch.
Hard-won lessons evaporate
The fix you found yesterday is gone today, and the agent repeats the same mistake.		 Lessons compound
Takeaways get written down as you go; small ones mature into durable, reusable runbooks the agent reaches for on its own.
No record of how — or why
Weeks later nobody can reconstruct the reasoning behind a decision — the context lived in a chat window that's gone.		 An auditable trail
Goals, decisions, and execution logs live in your project, surviving context resets and team hand-offs.
Knowledge is trapped in one silo
The working method lives in one person's head or one prompt history, so hand-offs are messy.		 Portable team knowledge
The method is plain-text files at the project root — readable by any developer or fresh agent.
Vendor and model lock-in
Switch model or tool and the context you built up in one tool's history is lost.		 Tool-agnostic
Plain text and folder conventions, so your method moves across any file-capable AI tool.
The human gets pushed out
Left unmanaged, agents run ahead on wrong assumptions and you lose control of what ships.		 You stay in the loop
You steer the spec, review progress, and sign off on what ships. Be the dotHuman in the loop.

Where to go next

  • New to the idea? Introduction — what it is, in about two minutes.
  • Want to see it work? Quickstart — a dotHuman on a real project in ~five minutes.
  • Want the mechanics? How dotHuman Works — the two ideas everything else builds on.

Related

Last updated