Impromptu UI
Generate interfaces for what you need to process or act on, instead of walls of text in chat. An agent's results, such as issues to triage or a Dependabot pull request to merge, appear as rows above the composer, one per item. Each row opens a side panel with the details, a draft you can edit, and buttons that post it, start a thread, run a command once you confirm it, or open a link. For design choices, a film strip shows screenshots of the original and each alternative, and Send feedback returns your pick and notes to the thread. The agent reads back what you did and carries on from there. Requires bb 0.43 or later, and needs no account or external service.
Overview
View sourceWhat you get
- Rows above the composer, one per item an agent wants you to act on, such as issues to triage or a Dependabot pull request to merge.
- A side panel for each item, with its details, a draft you can edit, and buttons that post the draft, start a thread, run a command once you confirm it, or open a link.
- A film strip for picking a design: screenshots of the original and its alternatives, a note on each, and Send feedback, which sends your pick and notes to the thread.
How it works
A skill writes its results as a view file and publishes it from the thread. Done and dismissed items sink to the bottom, and the agent reads back what you did. Views are stored in the plugin's own database on this machine.
For agents
Agents get the impromptu-ui skill and the bb impromptu-ui command: publish, state, hide, and list. The refine-impromptu-ui skill files an anonymized issue when a view falls short.
Requirements
Nothing to configure. A command button runs with your permissions, so read the command before you confirm it.
More from Daniel Bachhuber
More in Thread Content 56
Change what people see or do inside an open thread.



