We Turned Our Co-Prompting Protocol Into an Open Standard

What began as a verbal system inside PermitZIP is now a public draft for keeping human ownership clear when multiple AI agents are working in the same room.

The Open Co-Prompting Standard, Version 0.6 Draft, is now publicly available at co-prompting.com

Last week, I wrote about the verbal protocol our team created while running multiple AI agents in the same room.

We have now released that idea as The Open Co-Prompting Standard, Version 0.6 Draft.

The standard gives teams a shared protocol for assigning agent work during live meetings:

Call. Claim. Co-prompt. Close.

The standard organizes live agent work into four spoken beats: Call, Claim, Co-prompt, and Close.

It defines who owns each task, requires the prompt to be spoken aloud, allows others to add context directly, and forces every task to end with an explicit disposition: accepted, rejected, parked, or dropped.

The current draft includes 10 loop states, nine accountability rules, three conformance levels, an AEC profile, worked examples, and a one-page cheat sheet teams can use during meetings.

The AEC profile adds language for coordination, load, code, cut-sheet, drawing, constructability, value-engineering, pricing, RFI, and submittal work.

The standard is open, free, and licensed for others to adopt and adapt. It is also still a draft. We want practitioners to test it, challenge it, and help shape what comes next.

The site includes successful and failed meeting examples, including one where four accountability rules are broken without the room noticing.

Read the original article: The Next OAC Meeting Will Be a Room Full of People Prompting Agents

Read the proposed standard: co-prompting.com