Your agent reads the feedback, fixes the doc and closes the thread.

Every command answers in a format an agent can use: compact tables, exact follow-up commands, and a record of what it did and for whom. This site reads the same way: add .md to any address.

Comments as a table

Your agent pulls open comments as TOON, a table format that costs about half the tokens of JSON. Eight real comments came to 617 tokens as TOON and 1,250 as JSON.

$ cloudtifact comments --toon
comments[2]{id,thread,file,author,via,status,body}:
  77m5b9he,77m5b9he,PRDs/005-toon.md,gczh,null,open,"Add Claude's tokenizer numbers?"
  duqxuhwg,77m5b9he,PRDs/005-toon.md,gczh,claude-code,open,"Noted, next round."

A thread and its reply, one row each. Rows share the thread id.

Signed for the person it acted for

An agent acts as you, so its comments, replies, resolves and edits are yours. Each one also names the agent. Editing keeps every earlier version.

$ cloudtifact edit duqxuhwg "Noted: next round we count with Claude's tokenizer."
Edited duqxuhwg (version 2, via Claude Code).

$ cloudtifact history duqxuhwg
v1  2026-09-27 15:53 UTC  @gczh via Claude Code
v2  2026-09-27 16:06 UTC  @gczh via Claude Code

Set it up once

cloudtifact init --hooks adds a hook to Claude Code, Cursor or Codex, so docs publish the moment the agent saves them. cloudtifact skill install teaches agents the rest: read comments before editing, reply and resolve after, and share the link when they publish a page.