Tasks, actions, and notes

Tasks

Tasks is the queue of background work Chestnut does for you: writing conversation summaries (which is also when it looks for friction), the analysis you ask for with Analyze, suggesting fixes, writing up Notebook days, suggesting projects, and suggesting an arrangement for the session list. The first-run check that your background agent answers shows here too. The Type filter narrows the list to one kind.

Most of the time you can ignore it. It is worth opening when something you expected has not appeared yet, because a task that is still running looks different from one that failed. New conversations are imported separately; the status bar along the bottom of the window shows that.

Actions

Actions collects the fixes Chestnut has suggested for friction it found in your conversations. Where the Friction tab of a single conversation shows what went wrong in that one session, Actions is the view across all of them. Mark an action Complete or Dismiss it when you are done with it.

Acting on a suggestion opens an ordinary Chestnut session rather than an external terminal, so the work stays in your history like any other session. Launching from a suggested action also submits that action's prompt as the session's first turn, so the session starts with the problem already stated. Launching from a friction item instead opens a session at that conversation's directory and leaves the first turn to you.

If you want a plain shell instead, Open Terminal Here (under the arrow beside Launch Claude Code) opens your own terminal app in that folder and copies the action's prompt to the clipboard. After you use it, it becomes the main button for next time.

The Types tab holds the kinds of fix Chestnut suggests, each with the prompt behind it.

Working out a fix reads your project, but the agent runs in a folder Chestnut creates for it and is told where your project is. It can read your project from there but cannot change it: the only file it can write is its own answer, in its own folder. Text from your past conversations, which the suggestion is built from, cannot get it to plant a git hook, a plugin or agent settings in your project. On Claude Code, the background run also ignores your own Claude settings and permission flags, so nothing you allowed for your own sessions widens it. On Codex, it never shows its Folder access screen, and a repository's own Codex settings cannot take over the run.

Notebook

Notebook is a daily journal of your work. Chestnut writes one entry per day: what happened across every project, with line items linking back to the sessions behind them. You page through it a week at a time, or open the month to jump further, and a search box finds the day you are thinking of by anything written on it.

You can rewrite a day's entry with the Edit (pencil) button, add your own notes, or pin something from it; Pinned collects every pin. What you add is yours: re-writing the day's entry never overwrites your notes.

A day is written up the first time you open it, on the plan of the agent the Notebook runs on (your background agent, unless you picked another under Advanced in Settings > Analysis). A day with newer work offers an update, and only that re-writes it, so looking back at past days costs nothing. Nothing is written until you allow it in Settings > Notebook, where you can also turn the Notebook off.

Prompts

Prompts shows what Chestnut tells the AI, one row per feature. Sent into your sessions comes first and holds the session-start primer. Background tasks are the jobs Chestnut runs on its own, such as conversation summaries and the Notebook digest. Each row says when the feature runs. Your Quick Launch buttons are not listed here: you write and edit them in the Sessions pane.

Pick a feature to see What the AI receives, in order: its instructions, the input Chestnut fills in on each run, the output format, and where the result appears. Each part is marked Chestnut default, Your version, Filled in each run or Locked.

• Edit changes the wording of a feature's instructions. The parts Chestnut needs to read the answer, such as the JSON format, are locked: they show read-only where they sit, and you edit the text around them. Chestnut always sends its current version of a locked part, so a release that changes it still reaches your edited prompt.
• Compare with Chestnut's highlights the lines you added and removed.
• Restore default puts Chestnut's text back, including any later improvements to it.
• Changed by you lists every prompt and action type you have edited, with its date, and warns when Chestnut's default has changed since your edit. Action types are edited in Actions; the list links there.

Search covers feature names and the full text of every prompt.

How these fit together

These panels feed each other:

1. You run sessions in Terminal.
2. Chestnut summarises them in the background; that work shows in Tasks.
3. Friction it finds can be turned into suggested fixes in Actions.
4. Each day is written up in Notebook, where you add your own notes.
5. Prompts lets you adjust the wording Chestnut uses along the way.

You can use any part of it without the rest. Nothing here is required to run a session.

Where to go next

Agents driving Chestnut shows how the sessions you run can drive the app themselves.