daily-focus 1.2.0-dev.8 → 2.0.0-dev.14

package/README.md

@@ -35,9 +35,11 @@ To use your own data:
 npx daily-focus
 ```
 
-Until the first brief exists, the Today tab walks you through the rest: what it
-briefs, your objective, which coding-agent CLI writes the brief (Claude Code or
-Codex), and where it should look. Everything stays on the settings page afterwards.
+A new store opens on the setup screen: a few questions, one at a time, each with
+the likely answer already ticked. Once you've chosen the coding-agent CLI that writes
+the brief (Claude Code or Codex), it looks into the rest for you, such as which of
+your calendars are work ones, so most answers are a click. Everything stays on the
+settings page afterwards.
 To keep it running in the background on macOS, install it and run it as a service:
 
 ```sh
@@ -130,9 +132,20 @@ why is CI red, what does this thread actually want from me. With
 `DAILY_FOCUS_ASSISTANT` set to `claude` or `codex`, every item's panel gets an
 assistant that runs that CLI headless against the item, with quick actions for the
 common asks and a field for anything else. Click a card to open its panel, where its
-notes live too; the reply streams in there, and you can follow up in the same
-conversation. The panel stays on the item it was opened for while you move through
-the list.
+notes live too. While the assistant works, the panel shows what it is doing: each
+command it runs and each page, thread or ticket it reads, with a clock, and a cross
+on anything that failed, such as a repository it couldn't see. Once it answers,
+that trail folds away under the answer. The answer has buttons to copy it, or any
+block in it meant for pasting, and to open a draft it created. The field stays at
+the foot of the panel, so you can follow up in the same conversation. The panel
+stays on the item it was opened for while you move through the list.
+
+It can also answer before you ask. With **Answer ahead** on in the settings, when a
+pull request starts waiting on you because someone reviewed it or a check failed,
+the assistant assesses the review or explains the failure on its own. The first
+line of its answer appears on the row, such as "Flaky, not yours: the Safari job
+timed out", so the card already says what to do. It asks about each state once,
+one at a time, and no more than ten times a day unless you change the limit.
 
 It is told to read and draft, and nothing more. An email reply is created as a Gmail
 draft on the thread, so it is waiting there when you open the mail; a reply to a
@@ -150,8 +163,13 @@ in `DAILY_FOCUS_AGENT_AT`, and from the refresh icon beside the brief's age when
 morning's brief is out of date by lunch. Each run keeps what you have marked done,
 snoozed or noted.
 
+The agent reads your sources in parallel, handing each to a subagent of its own.
+While it runs, its panel shows what it is doing: what it says, each command it runs
+and each thread or page it reads, and each subagent under the step that started it,
+open while it works.
+
 Every run is recorded. Click "updated … ago" to open the latest: its report says what
-it wrote and what it could not reach, its working commentary is folded away beneath,
+it wrote and what it could not reach, how it got there is folded away beneath,
 and you can ask it why it did what it did, or to look again, in the same session. Earlier runs are folded away below. See
 [SETUP.md](SETUP.md#connect-your-morning-agent).
 

package/SETUP.md

@@ -27,18 +27,56 @@ and `npm start` runs it in the terminal. Each `daily-focus <command>` below is
 `npm run <command>` in a checkout, with two differences: `build-calendar` is
 `build:calendar`, and flags go after `--`, as in `npm run audit -- --against <file>`.
 
-Until the first brief exists, the Today tab walks you through setting it up:
-
-1. **What it briefs**: your work, or your personal life. See
+A new store opens on the setup screen. It first looks for Codex and Claude Code:
+on your login shell's PATH, then where their installers put them, and checks that
+the one it finds is logged in. With one ready it doesn't ask: it uses Codex if
+that is ready, otherwise Claude Code, and says so in a line you can **Change**.
+With neither ready, it says what is wrong with each and the command that fixes it
+(`codex login`, `claude auth login`), links to the installers, and takes the path
+to one it couldn't find.
+
+Then it asks a few questions, one at a time, and each comes with the answer you
+most likely want already ticked: confirm it or change it. A list can be left
+unticked and a field empty. Two questions can be left, by a button that says what
+that does: **Decide later** for the agent, and **Leave this out** for who you are.
+**Change** reopens an earlier answer.
+
+1. **Who writes the brief**, only when it couldn't pick for you.
+2. **What it briefs**: your work, or your personal life. See
    [A personal instance](#a-personal-instance) for running both.
-2. **Your objective**: the one thing you're trying to achieve, which the agent
-   ranks the day against.
-3. **Who writes the brief**: Claude Code or Codex, whichever it finds installed.
-4. **Where it should look**: your source list, which starts from a template with
-   what `gh` and `git` know about you filled in.
-5. **What else to show**: the pull request board, the Jira board and the live
-   agenda, each with its status and a button to turn it off.
-6. **Write my first brief**, which runs the agent once while you watch.
+3. **When it runs**, and on which days.
+4. **Who you are**: your name as meeting notes write it, your email, and each
+   GitHub account `gh` is logged in to, with what it's for.
+5. **What else to show**: the pull request board and the Jira board, each with
+   its status.
+6. **Which calendars** the brief reads, which the live agenda follows too.
+7. **Your objective**, which the agent ranks the day against, and what's in the
+   way. It suggests a few from what you've been doing, when something stands
+   out, and writes one up from whatever you tell it, however rough.
+
+The [assistant](#ask-the-assistant) isn't asked about: it comes on with the CLI
+that writes the brief, unless you've set it already.
+
+Once you've said what it briefs, the CLI looks into the others in the background,
+one short session each, so the slow ones are ready when you get to them. It reads
+`gh`, your mail and your calendars to tell work from personal, and says in a
+sentence why it ticked what it did. Reply under that sentence with what to change,
+such as "also the on-call rota", and it tries again, answering in the same place.
+Continue before it has finished and it asks first, then stops and keeps what the
+card holds.
+
+The sessions run on the newest flagship model at low effort (Opus for Claude Code,
+the first model in Codex's own list for Codex), which measured as fast as the
+lighter models and, unlike them, told your calendars from other people's. They
+only read, in an empty directory. Claude Code is given read tools by name, and
+under Codex every connector write is refused, drafts included; `gh` and `acli` are
+held to reading by their instructions, as
+[the assistant's](#what-the-assistant-can-do) are.
+
+Nothing is saved until the last step, **Write my first brief**. It saves the
+settings, writes your source list from the answers (unless you already have one,
+which it leaves alone) and your objective, restarts the dashboard to apply them,
+and runs the agent once while you watch.
 
 Everything in it stays on the settings page afterwards: the gear in the header, or
 <kbd>,</kbd>. The dashboard keeps its files in `~/.daily-focus/`, and links the
@@ -52,6 +90,9 @@ terminal open while you use the dashboard. `daily-focus --help` lists the rest.
 
 If you'd rather write the files by hand, `daily-focus init` creates the store with
 templates for both. It preserves existing files, so you can safely run it again.
+For a personal store, run `daily-focus init --profile personal`: it saves the
+profile in the store's settings, so the templates, the prompt and the dashboard
+agree.
 
 ## Set your objective and sources
 
@@ -91,24 +132,22 @@ Keep real names, account handles, and work details in this private store, outsid
 
 The dashboard runs the agent through a coding-agent CLI you already have. Install
 and log in to the [Codex CLI](https://developers.openai.com/codex/cli) or
-[Claude Code](https://code.claude.com/docs), and choose it in the setup steps or
-under **Morning agent** in the settings. In `.env` or the environment that is:
-
-```dotenv
-DAILY_FOCUS_AGENT=codex
-```
+[Claude Code](https://code.claude.com/docs), and choose it on the setup screen or
+as **Write the brief with** under **Morning agent** in the settings
+(`DAILY_FOCUS_AGENT`).
 
 Each scheduled morning at 07:00 your time the dashboard starts the agent through
 that CLI, headless, in the store, and the brief appears when it finishes. Set
-`DAILY_FOCUS_AGENT_AT` to another time, such as `06:30`. The dashboard has to be
+**Run it at** (`DAILY_FOCUS_AGENT_AT`) to another time, such as `06:30`. The dashboard has to be
 running at the time; if it wasn't, it runs the agent as soon as it is started that
 day, provided nothing has produced a brief for the day yet. It runs once a day,
 whatever became of the run: a failed morning is reported once, not retried every
 half minute. The days are Monday to Friday unless `DAILY_FOCUS_AGENT_DAYS` says
 otherwise; see [Match your working week](#match-your-working-week).
-`DAILY_FOCUS_AGENT_AT=off` means the dashboard never starts a run on its own; the
+**Run it at** set to `off` means the dashboard never starts a run on its own; the
 refresh icon still does. A new store's first brief always comes from the setup
-steps' button, so the clock never runs the agent before you've told it where to look.
+screen's last button, so the clock never runs the agent before you've told it where
+to look.
 
 "Your time" is the time zone of the browser you last looked at the dashboard
 in. The page tells the dashboard when it loads, reconnects or comes back into view,
@@ -172,18 +211,9 @@ agent to read.
 ## Configure integrations
 
 Settings are saved in the store, in `settings.json`, by the dashboard's settings
-page; saving one restarts the dashboard to apply it once nothing is running. The
-settings below can also go in a `.env` in the repo, which is how they were set
-before the store kept its own:
-
-```sh
-cp .env.example .env
-```
-
-A value saved in the store beats the same one in `.env`, and a variable set in the
-environment the dashboard starts with beats both; the settings page shows such a
-value but can't change it. `.env` is gitignored because account and organisation
-names belong outside git.
+page; saving one restarts the dashboard to apply it once nothing is running. Each
+is named below by its label on the page, and by its key in `settings.json`, in case
+you edit the file by hand; restart the dashboard after doing that.
 
 ### GitHub
 
@@ -194,22 +224,19 @@ gh auth login
 ```
 
 The board uses the active account by default. To choose accounts and repositories,
-set these in `.env`, replacing the invented examples with your own:
-
-```dotenv
-DAILY_FOCUS_GITHUB_ACCOUNTS=alice,alice_corp
-DAILY_FOCUS_GITHUB_SCOPE=acme,acme-labs/webapp
-```
+set **Accounts** (`DAILY_FOCUS_GITHUB_ACCOUNTS`), such as `alice, alice_corp`, and
+**Only these organisations or repositories** (`DAILY_FOCUS_GITHUB_SCOPE`), such as
+`acme, acme-labs/webapp`, under **Pull requests**.
 
 Named accounts stay fixed even if you switch the active account in another terminal.
 For organisations using SAML, make sure the account's token has access. The board
 reports access problems instead of silently showing no pull requests.
 
 If your repository has a check that represents its merge policy, list its exact,
-case-sensitive name in `DAILY_FOCUS_GITHUB_MERGE_GATE_CHECKS`. Separate names with
-commas. Pending matches appear under *Waiting on merge gate*.
+case-sensitive name in **Merge gate checks** (`DAILY_FOCUS_GITHUB_MERGE_GATE_CHECKS`).
+Separate names with commas. Pending matches appear under *Waiting on merge gate*.
 
-Set `DAILY_FOCUS_GITHUB=off` to disable the board.
+Turn **Pull request board** (`DAILY_FOCUS_GITHUB`) off to disable the board.
 
 ### Jira
 
@@ -219,14 +246,11 @@ Sign in with the Atlassian CLI:
 acli jira auth login
 ```
 
-The board uses that session. You can limit the projects and adjust which statuses
-need attention:
-
-```dotenv
-DAILY_FOCUS_JIRA_PROJECTS=PROJ,OTHER
-DAILY_FOCUS_JIRA_HOLD_STATUSES=Blocked,On Hold
-DAILY_FOCUS_JIRA_IN_PROGRESS_STATUSES=In Progress
-```
+The board uses that session. Under **Jira tickets** you can limit the projects with
+**Only these projects** (`DAILY_FOCUS_JIRA_PROJECTS`), such as `PROJ, OTHER`, and
+adjust which statuses need attention with **Statuses that mean on hold**
+(`DAILY_FOCUS_JIRA_HOLD_STATUSES`), such as `Blocked, On Hold`, and **Statuses that
+mean working on it** (`DAILY_FOCUS_JIRA_IN_PROGRESS_STATUSES`), such as `In Progress`.
 
 Hold statuses suppress *In flight with nothing linked* reminders. They don't hide
 other status mismatches. The in-progress list filters the *Working on* view;
@@ -239,10 +263,10 @@ for workflow limits and undo behavior.
 
 The board also asks GitHub about your pull requests, because Jira can learn of
 them days late. It uses the pull request board's accounts and scope, and nothing
-else needs to be configured. With `DAILY_FOCUS_GITHUB=off`, the ticket board uses
+else needs to be configured. With the pull request board off, the ticket board uses
 Jira alone.
 
-Set `DAILY_FOCUS_JIRA=off` to disable the board.
+Turn **Jira ticket board** (`DAILY_FOCUS_JIRA`) off to disable the board.
 
 ### Live calendar on macOS
 
@@ -254,19 +278,16 @@ goes in `~/Library/Application Support/Daily Focus/`, where every copy finds it:
 daily-focus build-calendar
 ```
 
-Choose calendar names exactly as Calendar.app shows them, and list your own email
-addresses so the dashboard can recognize meetings you've declined:
-
-```dotenv
-DAILY_FOCUS_CALENDARS=Work,Personal
-DAILY_FOCUS_CALENDAR_ADDRESSES=you@example.com,you@personal.example
-```
+Tick the calendars under **Calendar** in the settings (`DAILY_FOCUS_CALENDARS`,
+spelled exactly as Calendar.app shows them), and list your own email addresses in
+**Your email addresses** (`DAILY_FOCUS_CALENDAR_ADDRESSES`) so the dashboard can
+recognize meetings you've declined.
 
 Start the dashboard and allow the helper's calendar access when macOS asks.
 Rebuilding the helper may prompt again.
 
 Without live calendar access, the agenda uses events from the morning brief.
-Set `DAILY_FOCUS_CALENDAR=off` to use those events explicitly.
+Turn **Live agenda** (`DAILY_FOCUS_CALENDAR`) off to use those events explicitly.
 
 ## Ask the assistant
 
@@ -279,12 +300,8 @@ is failing, what a thread is asking of you, a reply to an email drafted straight
 Gmail as a draft on the thread.
 
 Install and log in to [Claude Code](https://code.claude.com/docs) or the
-[Codex CLI](https://developers.openai.com/codex/cli), then choose it under
-**Assistant** in the settings, or set:
-
-```dotenv
-DAILY_FOCUS_ASSISTANT=claude
-```
+[Codex CLI](https://developers.openai.com/codex/cli), then choose it as **Answer
+questions with** under **Assistant** in the settings (`DAILY_FOCUS_ASSISTANT`).
 
 Once the dashboard has restarted, the assistant appears in the panel. Model and effort default to whatever
 the CLI is configured with; `DAILY_FOCUS_ASSISTANT_MODEL` and
@@ -314,10 +331,11 @@ instructions. What the CLI itself enforces:
   `mcp__claude_ai_Gmail` allows every tool on the Gmail connector, sending and
   deleting included, and `gh` and `acli` can comment, merge and move tickets as well
   as read. Only the instructions keep it from doing those. To hold it to reading
-  mail and drafting replies, name the Gmail tools instead of the connector:
+  mail and drafting replies, name the Gmail tools instead of the connector, in
+  **Tools Claude Code may use** under **Assistant** (`DAILY_FOCUS_ASSISTANT_TOOLS`):
 
-  ```dotenv
-  DAILY_FOCUS_ASSISTANT_TOOLS=Bash(gh *),Bash(acli *),WebFetch,mcp__claude_ai_Gmail__search_threads,mcp__claude_ai_Gmail__get_thread,mcp__claude_ai_Gmail__get_message,mcp__claude_ai_Gmail__list_labels,mcp__claude_ai_Gmail__list_drafts,mcp__claude_ai_Gmail__get_draft,mcp__claude_ai_Gmail__create_draft,mcp__claude_ai_Gmail__update_draft
+  ```text
+  Bash(gh *),Bash(acli *),WebFetch,mcp__claude_ai_Gmail__search_threads,mcp__claude_ai_Gmail__get_thread,mcp__claude_ai_Gmail__get_message,mcp__claude_ai_Gmail__list_labels,mcp__claude_ai_Gmail__list_drafts,mcp__claude_ai_Gmail__get_draft,mcp__claude_ai_Gmail__create_draft,mcp__claude_ai_Gmail__update_draft
   ```
 
   A tool left off the list is refused, so if the connector renames one, the
@@ -339,6 +357,26 @@ instructions. What the CLI itself enforces:
   assistant or the morning agent on it at all; before 0.149.0 it can't switch
   computer use off, so the assistant won't answer while that setting is off.
 
+### Answer ahead
+
+With **Answer ahead on pull requests** (`DAILY_FOCUS_ASSISTANT_AUTO`) on, the
+dashboard asks the assistant before you do. When a pull request lands in *Waiting on
+you* because someone reviewed or commented, it asks the assistant to assess the
+review. When a check fails, it asks for an explanation of the failure. The first
+line of the answer appears on the row,
+on the board and on the brief item for the same pull request, while it still
+matches the pull request's state. The whole answer is in the panel, as a turn you
+can follow up on.
+
+It asks about each state once. A new comment or a new failure is a new question;
+the same checks failing again on the same commit are not. It runs one question at a
+time and never on a row you're asking about, and it starts no more than
+`DAILY_FOCUS_ASSISTANT_AUTO_LIMIT` a day (10 by default). It uses only those two
+quick actions, which read and answer, and never one that drafts. It follows the
+board, which reads only while a tab is open, so it asks while you have the
+dashboard open. It is off by default because it spends the CLI's usage without
+you pressing anything.
+
 Its instructions live in `prompts/assistant.md`, linked into the store as
 `assistant.md` each time the dashboard starts. What it says stays in the chat in the panel: it
 never leaves notes on the item or marks it handled. That is yours to do once you have
@@ -354,10 +392,17 @@ in the archive. While it runs, a banner says so and offers Stop; click the banne
 to watch the run. When it finishes the banner goes, since the brief it wrote is the
 page. A run that fails stays as a warning until you dismiss it.
 
+The agent reads your sources in parallel, each handed to a subagent of its own, and
+the panel shows what it is doing as it goes: what it says, each command it runs and
+each thing it reads, and each subagent under the step that started it, open while
+it works. Claude Code reports its subagents itself. Codex doesn't, so for a Codex
+run the dashboard reads them from Codex's own session files, under `~/.codex/sessions`
+or `$CODEX_HOME`. It only ever reads them.
+
 Click "updated … ago" to open the latest run in the panel. It holds the agent's
 report — what it wrote, what it dropped as handled, and any source it could not
-reach — with everything it said on the way there folded away beneath, and a field
-to ask it about the run. The question goes to the agent in the
+reach — with everything it said and did on the way there folded away beneath, and a
+field to ask it about the run. The question goes to the agent in the
 same session, so it answers from what it actually did: why it dropped an item, what
 it could not find, whether it can find it now. A follow-up is given nothing to
 write; if something in the brief should change, it says what, and you run it again.
@@ -429,10 +474,11 @@ checkout it rebuilds the page each time the dashboard starts. It won't install f
 The LaunchAgent is written for this machine and isn't tracked in git, so each
 machine runs the command once. It uses the Node.js that ran it and keeps two things
 from your shell: `PATH`, so the dashboard finds `gh`, `acli` and your agent's CLI,
-and any `DAILY_FOCUS_` variables set there. Each time it starts, the dashboard also
-asks your login shell for its `PATH`, so a CLI installed since is found without
-reinstalling the service. Everything else comes from the store's settings and
-`.env`, which are read each time the dashboard starts. A CLI that relies on a token
+and the store, port and host, whether given as `--data`, `--port` and `--host` or set
+as `DAILY_FOCUS_DATA`, `DAILY_FOCUS_PORT` and `DAILY_FOCUS_HOST`. Each time it starts,
+the dashboard also asks your login shell for its `PATH`, so a CLI installed since is
+found without reinstalling the service. Everything else comes from the store's
+settings, which are read each time the dashboard starts. A CLI that relies on a token
 exported in your shell profile, rather than its own login, won't find it.
 
 Each store gets one service. Running the command again for the same store replaces
@@ -451,19 +497,15 @@ daily-focus service --remove
 
 Daily Focus can brief your personal life as well as your work: email, family
 calendars, Apple Reminders, Apple Messages, e-Boks and your own GitHub projects. Run it as a second instance
-with its own store, server and morning agent, and choose **Personal life** in its
-setup steps, or set:
-
-```dotenv
-DAILY_FOCUS_PROFILE=personal
-```
+with its own store, server and morning agent, and choose **Personal life** on its
+setup screen (`DAILY_FOCUS_PROFILE`).
 
 The dashboard then links the personal prompt, `prompts/morning-brief-personal.md`,
 into the store as `prompt.md`, and the source list starts from a template with
 sections for those sources. The profile also switches off
 the Jira board, and gives the tab a green house icon instead of the blue briefcase. The pull request
 board stays on for side projects; point it at your personal account with
-`DAILY_FOCUS_GITHUB_ACCOUNTS`, or set `DAILY_FOCUS_GITHUB=off`.
+**Accounts**, or turn the board off.
 
 The personal sources often live on another machine, such as a home Mac with access
 to Messages and Reminders. Run the instance there, keep `DAILY_FOCUS_HOST` on
@@ -482,40 +524,45 @@ Tailscale `ts.net` name, or by the name in `DAILY_FOCUS_HOST`. That stops a web
 from pointing its own name at your dashboard and reading your brief. Reached any
 other way, it replies `403` and names the host it was reached as.
 
-To run both instances on one machine, start the second with its own
-`DAILY_FOCUS_DATA`, `DAILY_FOCUS_PORT` and `DAILY_FOCUS_PROFILE` in its environment,
-which takes precedence over `.env`. `daily-focus service` keeps those variables, so this
-installs the second service beside the first. Use the same variables each time you
+To run both instances on one machine, give the second its own store and port.
+`daily-focus service` keeps them, so this installs the second service beside the
+first, and its setup screen asks what it briefs. Use the same flags each time you
 run it for that instance:
 
 ```sh
-DAILY_FOCUS_PROFILE=personal DAILY_FOCUS_DATA=~/.daily-focus-personal DAILY_FOCUS_PORT=4322 daily-focus service
+daily-focus service --data ~/.daily-focus-personal --port 4322
 ```
 
 ## Match your working week
 
-Set `DAILY_FOCUS_AGENT_DAYS` to the days the dashboard runs the agent. It uses cron
-weekday numbers: `1-5` for Monday to Friday, `0-4` for Sunday to Thursday, or `0,6`
-for weekends. Unset means Monday to Friday. `daily-focus audit` shows the schedule in
-use. Days without scheduled runs don't make a brief overdue. Once a refresh is due,
-the dashboard allows 45 minutes before showing a warning.
+Set **On these days** under **Morning agent** (`DAILY_FOCUS_AGENT_DAYS`) to the days
+the dashboard runs the agent. It uses cron weekday numbers: `1-5` for Monday to
+Friday, `0-4` for Sunday to Thursday, or `0,6` for weekends. Unset means Monday to
+Friday. `daily-focus audit` shows the schedule in use. Days without scheduled runs
+don't make a brief overdue. Once a refresh is due, the dashboard allows 45 minutes
+before showing a warning.
 
 ## Settings reference
 
-All settings can be changed on the settings page, which saves them in the store,
-or set in `.env` or the environment; restart after changing those two by hand.
-`DAILY_FOCUS_DATA`, `DAILY_FOCUS_PORT` and `DAILY_FOCUS_HOST` say where the store is
-and how the dashboard listens, so they are never kept in the store and only the
-environment or `.env` can change them. Their defaults suit almost everyone.
-The brief's `dayStart` and `dayEnd` override the configured working hours. Neither
-matters when `DAILY_FOCUS_FREE_WINDOWS` is off.
+Three settings say where the store is and how the dashboard listens, so they are
+never kept in the store. Give them as flags, or set them in the environment the
+dashboard starts in; a flag wins. Their defaults suit almost everyone.
+
+| Flag | Variable | Default | Purpose |
+|---|---|---|---|
+| `--data` | `DAILY_FOCUS_DATA` | `~/.daily-focus` | Where the store lives |
+| `--port` | `DAILY_FOCUS_PORT` | `4321` | Port for the dashboard |
+| `--host` | `DAILY_FOCUS_HOST` | `127.0.0.1` | Loopback only by default, since this is personal data. A name here is also one the dashboard answers to, besides localhost, IP addresses and `ts.net` names |
+
+Everything else is set on the settings page, which saves it in the store's
+`settings.json` under the keys below; restart the dashboard after editing that file
+by hand. Nothing else is read from the environment. The brief's `dayStart` and
+`dayEnd` override the configured working hours. Neither matters when
+`DAILY_FOCUS_FREE_WINDOWS` is off.
 
-| Variable | Default | Purpose |
+| Key | Default | Purpose |
 |---|---|---|
 | `DAILY_FOCUS_PROFILE` | `work` | `work` or `personal`. Picks the prompt the dashboard links and the defaults marked below |
-| `DAILY_FOCUS_DATA` | `~/.daily-focus` | Where the store lives |
-| `DAILY_FOCUS_PORT` | `4321` | Port for the dashboard |
-| `DAILY_FOCUS_HOST` | `127.0.0.1` | Loopback only by default, since this is personal data. A name here is also one the dashboard answers to, besides localhost, IP addresses and `ts.net` names |
 | `DAILY_FOCUS_FREE_WINDOWS` | `on`; `off` for personal | `off` shows events only: no free windows in the agenda, and the next event in place of focus time left |
 | `DAILY_FOCUS_WORK_START` | `9` | Local hour the working day starts, when the brief doesn't say |
 | `DAILY_FOCUS_WORK_END` | `17` | Local hour the working day ends when the brief does not specify it |
@@ -546,11 +593,14 @@ matters when `DAILY_FOCUS_FREE_WINDOWS` is off.
 | `DAILY_FOCUS_ASSISTANT_EFFORT` | *the CLI's own* | Effort passed to the CLI untouched: `low`, `medium`, `high`, `xhigh` or `max`. Unset passes no flag |
 | `DAILY_FOCUS_ASSISTANT_TOOLS` | `Bash(gh *),Bash(acli *),WebFetch,mcp__claude_ai_Gmail` | What Claude Code may use without asking, in its permission syntax, comma-separated. Editing tools are denied regardless. The default allows every Gmail tool, sending included; see [What the assistant can do](#what-the-assistant-can-do) to narrow it. Ignored by Codex |
 | `DAILY_FOCUS_ASSISTANT_COMPUTER_USE` | `off` | `on` lets the assistant use Codex's computer-use and browser plugins, which act in your apps and browser with your sign-ins. `off` switches them off for its runs. Ignored by Claude Code, which uses only the tools in its list |
+| `DAILY_FOCUS_ASSISTANT_AUTO` | `off` | `on` asks the assistant ahead about pull requests waiting on you, after a review or a failing check, and puts the answer's first line on the row. See [Answer ahead](#answer-ahead) |
+| `DAILY_FOCUS_ASSISTANT_AUTO_LIMIT` | `10` | How many times a day the assistant may start on its own, from 1 to 100 |
 | `DAILY_FOCUS_AGENT` | `off` | `codex` or `claude` lets the dashboard run the morning agent through that CLI: each scheduled morning, and from the refresh icon beside the brief's age. `off` hides it |
 | `DAILY_FOCUS_AGENT_AT` | `07:00` | Local time, `HH:MM`, at which the dashboard starts the agent on each scheduled day, or as soon as it is running after that. `off` means it never starts one on its own; the refresh icon still does. Needs `DAILY_FOCUS_AGENT` |
 | `DAILY_FOCUS_AGENT_BIN` | *the CLI's name* | Path to that CLI, for when the server's PATH lacks it. `~` is expanded |
 | `DAILY_FOCUS_AGENT_MODEL` | *the CLI's own* | Model passed to the CLI untouched. Unset passes no flag |
 | `DAILY_FOCUS_AGENT_EFFORT` | *the CLI's own* | Effort passed to the CLI untouched. Unset passes no flag |
+| `DAILY_FOCUS_AGENT_SEARCH_TRASH` | `off` | Whether the agent searches the mail account's Trash as well as the inbox and archive. The dashboard writes it to `agent-settings.json` in the store before each run, where the agent reads it |
 | `DAILY_FOCUS_AGENT_TOOLS` | `Bash(gh *),Bash(acli *),WebFetch,mcp__claude_ai_Gmail` | What Claude Code may use to reach your sources, in its permission syntax, comma-separated. Writing `items.json`, and nothing else in the store, is allowed regardless. The default allows every Gmail tool, sending included. Ignored by Codex |
 
 ## If something looks wrong
@@ -559,9 +609,10 @@ matters when `DAILY_FOCUS_FREE_WINDOWS` is off.
   which it ends with. Run `daily-focus service` to keep it running. If the
   service is installed and the page still doesn't load, read
   `~/Library/Logs/local.daily-focus.log`.
-- Before the first brief, the Today tab shows the setup steps. Once a brief has
-  existed, an empty Today tab means the agent hasn't written a new `items.json`:
-  start a run from the refresh icon beside the brief's age.
+- After setup and before the first brief, the Today tab says where the first brief
+  is, and offers to run it again if it didn't finish. Once a brief has existed, an
+  empty Today tab means the agent hasn't written a new `items.json`: start a run
+  from the refresh icon beside the brief's age.
 - A stale brief needs an agent run. Refreshing a board doesn't regenerate the brief;
   the icon beside the brief's age does. Click "updated … ago" to see what the last
   run said, or whether there was one: the dashboard has to be running at

package/dist/{audit-DNTFZOHH.js → audit-AQUBVWPA.js}

@@ -1,7 +1,7 @@
 import {
   readSystemZone,
   startingZone
-} from "./chunk-6SMUSPZG.js";
+} from "./chunk-YPGKXTBX.js";
 import {
   linkState
 } from "./chunk-QKT2FAB5.js";
@@ -13,7 +13,8 @@ import {
   resolveItems,
   runContractChecks,
   setsStatus
-} from "./chunk-GQ3NCWRT.js";
+} from "./chunk-ITUK7XKS.js";
+import "./chunk-QG4XXKBP.js";
 import {
   BRIEF_REFRESH_GRACE_HOURS,
   describeSchedule,
@@ -23,7 +24,7 @@ import {
   resolveSchedule,
   runsOn,
   workingMsBetween
-} from "./chunk-KZNY45XW.js";
+} from "./chunk-72NUDSOV.js";
 import {
   command
 } from "./chunk-BC2W7NND.js";