Comment on what your agent built. It reads the thread and fixes it.

A short one from Carl and me. Reviewing what an agent makes has always had a gap: you read the page, you have notes, and the notes land somewhere the agent can’t see.

Usually that feedback goes into a Slack message, a quick “can you change the second chart,” or a fresh prompt that has lost track of which paragraph you meant. The page and the comment on it live in two different places.

So we closed the gap. You can now leave inline comments on any published artifact, and the agent that built it can read them and act.

What shipped

Select text on a published page and leave a comment anchored to that exact spot, the way you would in Google Docs. You get threads, one level of replies, resolve and reopen, and an All Discussions panel with Open and Resolved tabs. It works on both HTML and Markdown artifacts.

Comments stay put across new versions, as long as the text they point at still exists. Edit that text away and the comment moves to the panel marked “context unavailable in this version,” with a link back to the version it was written on. Nothing silently disappears.

The part we actually care about

The comment is not just for the next human. The agent that made the page can read the thread and respond. Point it at the artifact and it reads the open comments, makes the edit, replies on the thread and resolves it.

# your agent, over MCP or the CLI
dsp comment list --artifact oXpTBjGg --status open
dsp comment add  --artifact oXpTBjGg --parent <commentId> --body "Fixed – pulled the Q3 figures from the latest export."
dsp thread resolve <commentId>

Or in plain language, with the display.dev skill installed:

# in Claude Code
"Read the open comments on this artifact and address them."

Because every comment is signed with who left it, the agent can tell a human’s directive from its own earlier note and weight them accordingly. (More on signed identity in a companion post.)

Worth knowing

• Comments are for your team. Anyone signed in with access in your organisation can comment; anonymous visitors just see the page.
• Comments are free on every tier.
• Threads are anchored to the text, so they follow the content rather than a line number.

Where it could be handy

• Reviewing a report or dashboard an agent generated, without copying notes into Slack.
• Leaving “tighten this, cut that” directly on a draft and letting the agent work through the thread.
• A reviewer and an agent going back and forth on the same page until it is right.