Copycat by 29Court

Your rules outlive your agent.

A new agent shouldn't meet you as a stranger.

A rule is a correction kept in words you can read and edit.

A correction is you saying what was wrong, once.

An agent swap is the assistant changing while the rules stay.

01

The difference rules make

Demonstration. The example below is invented to show how a correction lands. It is not a real client correction.

“Never start my draft with ‘Great question.’ Just answer.”

New agent, no rules Great question. Happy to help with that draft. Before we start, could you tell me how you like your openings?
Agent carrying your rules Here is the draft, opening on the point itself.

The rule it became: Open with the answer. No greeting filler, no asking for a preference already taught.

02

How a correction becomes a rule

Teach You correct the agent in your own words, or say what right looks like before it starts. Mechanism: Bot Manners keeps the correction as a trace, in your words, with the rule it becomes.
Correct The correction becomes a rule: visible words you can read, edit, or remove, kept with the trace it came from. Mechanism: the trace becomes a rule on your Bot Manners rules page, kept under your enrollment.
Serve When an enrolled agent asks the Rules MCP, it gets the bundled rule corpus fresh, not a stored copy. Mechanism: the Rules MCP serves the bundled rule corpus fresh on each request, read-only, under your enrollment key.
A new agent inherits You hand a new agent your rules address and enrollment. It reads your rules before it works, and it does not meet you as a stranger. Mechanism: you hand it over in that order: identity first, then your rules, then the tools it may use.

Its next correction goes back to the first step, into the same rules.

An illustration that runs on this page only. Nothing you type is sent or stored, and it does not teach a real agent. On this page your words are kept exactly as you wrote them. Your real rules live on your Bot Manners rules page, under your enrollment.

Rules made on this page

  • No rules yet on this page.
03

The four parts

The rules engine Keeps your corrections as rules you can read and edit, on a page opened by your key. Mechanism: your enrollment, and the rules service at rules.29court.com.
Bot Manners Where you give a correction, and where a draft you send is checked before it goes out. Mechanism: your words, kept as a correction trace with the rule they become.
Dillo The backend your accounts and agents run on: sign-in, sessions, and one named home per client. Mechanism: accounts and passkeys, with tenancy by named home.
Dillo Flows The execution layer for work that runs in steps and can wait on a human. Mechanism: steps that can pause for a human decision. Named flows are future until one is running and probed. Future parts
04

The connect path

Connect an agent to the Rules MCP

Add that endpoint as an MCP server in an agent that takes one, with your enrollment key. The agent can ask for the bundled rule corpus at the start of a session, and check a draft against it. Your own rules live on your Bot Manners rules page; that address is what you hand a new agent.

The round trip, in order

  • An agent connects to mcp.29court.com and presents its enrollment key.
  • The agent asks for the rules with get_rules.
  • The Rules MCP returns the bundled rule corpus, in words you can read.
  • A draft sent with check_draft comes back checked against those rules.
  • The check names each rule the draft broke, with a fix where one is known.
  • The drafts in this sequence are examples. No private draft or correction is shown on this page.

    What this does not prove: it does not prove another client's rules, and it makes no claim that this page's example is live. The endpoint above is the real connect path for the Rules MCP.

    Swap the assistant. Keep what you taught it.

    Change the agent. Keep the rules.

    Start with Bot Manners