Agents can now set an expiry date and draw on Whiteboards
1 day ago
This pull request teaches agents two new things about Artifacts. They can make one expire on a date, and they can work on a Whiteboard alongside people. Both arrive as new text in the skill file that agents read before they touch an Artifact. The pull request body is just its title, so everything here comes from the diff.
Ask about this presentation
Chapters
Show transcriptHide transcript
This pull request teaches agents two new things about Artifacts. They can make one expire on a date, and they can work on a Whiteboard alongside people. Both arrive as new text in the skill file that agents read before they touch an Artifact. The pull request body is just its title, so everything here comes from the diff.
The whole change sits in one markdown file, SKILL.md, in the artifacts skill. It adds twenty lines and replaces one. Expiry goes into the delete section, Whiteboards becomes a new section eight, and the old Pin heading is the single replaced line, because Pin moves down to section nine.
When a user wants something gone later, the skill now tells the agent how to say so. Pass expires_in_days on create or update, in an update call that has no content. Or pass expires_at for a set moment. Setting expires_at to null cancels it. The examples in the skill are keeping something for a month, or a poll that runs a week.
The skill says an expired Artifact is deleted the same way delete_artifact deletes it. Within a day of the date, it moves to the owner's Trash. It sits there for 30 days. After that, every version, every comment and every file is gone. The owner gets a month to change their mind.
Deleting data is the risky part, so the skill adds guardrails. An agent sets an expiry only when the user asks, and the skill says that without one an Artifact is kept. Only the owner can set or cancel an expiry. And whenever the agent sets one, it tells the user the date.
Whiteboards are a new kind of Artifact. An agent makes one with create_artifact, and gives it no content. The board.shapes field starts it off with frames, stickies, boxes and arrows. The spec value that marks it as a Whiteboard is bisque-canvas slash v1.
Two tools do the work. The read board tool returns the board as structure, with its frames, its shapes and what each arrow connects. The skill says to read it before changing anything or answering a comment about it. The edit board tool adds, moves, restyles or removes shapes, keyed by ids the agent chooses. It only touches shapes the agent drew. Shapes people drew are theirs, so the agent reads them and leaves them alone.
There are no before and after numbers here, because the change is instructions for agents rather than code that runs. The author's description reads, update artifacts skills, followed by a pointer to the contributing guide. The pull request doesn't include tests and has no review comments, and it hasn't merged. So the question to settle is whether the new text matches how the tools really behave.
Since the change is only text, review it as text. Check the expiry numbers and the Trash window against what the service does, and check that the Whiteboard tools behave as described. Once it merges, agents can set an expiry date and draw on Whiteboards. The pull request is number six in the artifacts repository.
Up next
2:52Artifact skills now tell agents how to fix a broken Bisque sign-in
@bisque-cloud/artifacts
2 days ago
2:39Agents now publish from the account that owns your work instead of asking
@bisque-cloud/artifacts
2 weeks ago
3:28Your agent's answer becomes a page at a link that never changes.
@bisque-cloud/artifacts
2 weeks ago
3:00Pull requests now get explained, when one is worth it
@bisque-cloud/artifacts
2 weeks ago









