Agent Revision Markup
Track changes for agents.
Agent Revision Markup defines how agent document actions are recorded inside a DOCX.
The human surface stays familiar: comments, tracked changes, and ordinary document text. The agent surface is structured: a signed record that says who acted, under what authority, what changed, and what a human approved.
It does not sign the agreement. It signs the record of actions around the agreement.
Why it exists
Documents move between tools. The record of agent work should move with them.
Without a shared format, one system's agent history becomes another system's unreadable platform log. Agent Revision Markup gives compatible tools the same basic facts:
- who acted
- what changed
- under what authority
- what a human approved
- where that revision applies in the document
Agent context across interfaces
Websites can publish llms.txt so models have a predictable place to find LLM-friendly site context. Repositories can publish AGENTS.md so coding agents have a predictable place for project instructions.
DOCX files need their own agent-readable context layer too. Agent Revision Markup is that layer for documents. It is not an instruction file for an agent to obey. It is the revision record: who acted, under what authority, what changed, what a human approved, and which tool can continue the work.
Start from the website, open the demo app, or continue with what is stored in the DOCX.