Update artifacts skills
2 hours ago
This pull request adds two Agent Skills to bisque-cloud artifacts. Ask your agent for a plan, a report, or a sign-up sheet, and it publishes the answer as a page at a link that never changes. When the plan changes, the agent updates the same link, and when people comment, it answers them.
Ask about this presentation
Chapters
Show transcriptHide transcript
This pull request adds two Agent Skills to bisque-cloud artifacts. Ask your agent for a plan, a report, or a sign-up sheet, and it publishes the answer as a page at a link that never changes. When the plan changes, the agent updates the same link, and when people comment, it answers them.
The README opens with what you can ask for. Write up the migration plan as an Artifact and share it with Ana as an editor. Your agent writes the document, publishes it, shares it, and hands you the link. Ana can open it, comment on it, and publish changes of her own.
The skills split the work in two. The artifacts skill publishes a Markdown document or an HTML page, updates it at the same link, shares it, and deletes it. The artifact-comments skill finds the threads that mention the agent, answers them, and resolves them. Both call the Bisque MCP server, so nothing is stored locally.
To publish, the agent calls create_artifact. It sends a title, a type, and the complete content. Markdown suits a document read top to bottom, like a plan or meeting notes. HTML suits a dashboard, a form, or a poll, and it can keep shared data that everyone who opens the page sees live.
Updating works the same way. The agent calls update_artifact with the artifact ID and the whole new content. That publishes a new version at the same URL, and the earlier versions stay in the history. The skill is explicit that the agent never creates a second Artifact to update the first, because the link you already sent would go stale.
Who can open the page is the audience field. It starts at owner, so only you. Users opens it to the people you name, and org opens it to a bisque.team company. Public means anyone with the link. That deletes every comment thread, because comments don't exist on a public page, so the skill refuses unless the user has said in words that the comments can go.
The second skill closes the loop. It lists the open threads where someone mentioned the agent. It makes the change and republishes. It replies in the thread with what changed and in which version. Then it resolves the thread. The order matters, because a thread resolved without a reply leaves the person guessing.
CONTRIBUTING calls the rules at the end of each SKILL.md load-bearing rather than stylistic. The agent confirms before anything that removes access or content. It never guesses an email address, since sharing with a guess shares with whoever owns it. And it updates the existing Artifact instead of publishing a copy. A change that removes one has to say what replaces it.
The pull request body is two lines, so there is no benchmark or before and after to show, and the review thread has no comments. That is the whole description. What changed is six files: a plugin manifest at version 1.0.0, CONTRIBUTING, an MIT license, the README, and the two SKILL.md files. So the useful review is of the wording, because the SKILL.md files are the instructions your agent will follow.
To try them, run this one command. It copies both skills into your agent's skills directory. The limit is that they need a Bisque account and network access to bisque.today, because everything runs through the MCP server. Then your agent's answer becomes a page at a link that never changes.
