Learning Exercises
Get a short exercise on the code your agents just wrote, so you keep understanding what lands in your repository. After a commit, new files, a schema change or a merged pull request, one line above the composer offers a 10–15 minute exercise; offers are decided from bb's own data, not by a model, and are capped at three answered a day. Start runs the session in a side thread forked from yours, where each question is a card for your answer and how sure you are, and the agent's turn ends while the card is open, so it cannot hint. A Learning tab in every thread's side panel shows the session in progress, starts an exercise on the last commit, the new files or a topic you type, and keeps the project's paused and earlier sessions with your answers. A session runs on the thread's own agent and uses its tokens; answers are stored on the machine running bb. Exercises adapted from Dr. Cat Hicks's Learning Opportunities (CC BY 4.0), not affiliated with or endorsed by the author. Requires bb 0.48.1 or later with bb's bundled Git plugin on; pull request offers also need GitHub PR Status.
Overview
View sourceKeep learning the code your agents write. After a commit, new files, a schema change or a merged pull request, Learning Exercises offers a short exercise on that work, and runs it in a side thread with a real answer card.
What you get
- An offer at the right moment. When a turn ends after real work, one line appears above the composer naming what was done, with Start, Later and Not in this thread. Offers are decided from bb's own data (commits, the turn's file changes, pull request status), never by an AI, so they cost no tokens.
- Nothing in your agents' context until you start. The plugin ships no skill and adds no tools or instructions to your threads. Only a session you start receives the exercise instructions.
- Sessions beside your work. Start forks the thread, so the session's agent knows what was just built and your work thread is left as it was. Agents that cannot fork run it in the same thread.
- A real pause. Each question is a card: your answer, how sure you are from 1 to 5, Make it more specific, Skip and Pause. The agent's turn is over while the card is open, so it cannot hint. Your answers stay in the transcript in your own words, and the agent's feedback on each answer shows in the chat as a card with its verdict, and above the next question, so it never folds away.
- Answer forms. A question with several parts can come with a form instead of one text box: text, code, numbers, sliders, choices, yes/no, a scale, ordering, matching, lists and tables.
- The Learning tab. In each thread's side panel: the offer or the session in progress, with Pause, Resume and Finish; start an exercise on the thread's last commit, its new files or a topic you type; and the project's paused and earlier sessions with your answers.
- Caps that hold. At most 3 answered offers a day, 2 hours apart, 2 sessions per thread; Later snoozes a thread; three Laters in a row keep only commit and pull request offers for a week. Offers you leave unanswered expire without counting.
Where offers appear
Only in top-level threads, including ones another plugin starts for you, such as a GitHub board card's: never in child threads, automations (or other plugins you list in the settings), sessions' own threads or hidden workers. Turn offers off per thread from the thread menu, per project from the offer, or everywhere in the settings. In a sub-agent thread the Learning tab is still on the side panel's tab bar, to start an exercise on that thread's work by hand.
Requirements
- bb 0.48.1 or later.
- bb's bundled Git plugin on, to see commits; GitHub PR Status for pull request offers.
- Your answers are kept on this machine, in the plugin's own database, and can be deleted per session.
Credits
The exercises, the pause rule and the learning science are adapted from Dr. Cat Hicks's Learning Opportunities (https://github.com/DrCatHicks/learning-opportunities), CC BY 4.0. Not affiliated with or endorsed by the author.
More from Gustavo Ambrozio
More in Thread Content 83
Change what people see or do inside an open thread.



