AW-19 — Home, the morning screen

Behaviour-first spec per Constitution Principle IX. This document describes what a user sees and can do. No class names, no file paths, no code — those live in plan.md.

Feature ID: aw-19-home Program: Agent Workspace Branch: feat/aw-19-home Status: Draft Created: 2026-09-06 Last updated: 2026-09-06 Size: M · Blocking dependencies: AW-02, AW-03, AW-04 Extends: the dashboard home route (existing) · Adjacent epics: AW-01 Command palette, AW-09 Runs & receipts, AW-10 Schedules & calendar, AW-17 Costs & caps, AW-20 Onboarding

Additive by default (program rule #1). Nothing here deletes a block, a route, an endpoint or an i18n key. Every section that renders on Home today still renders after this epic — the inventory blocks (stats strip, Missions preview, Ideas, recent Works, Tasks, Agents) move below the new morning stack, and the two signal blocks that already exist (needs-attention, coming-up) are widened in place rather than replaced. One new entity is introduced (a per-user Home preference row) and is justified in §5.4.

Dependency posture. Home is a reader. It composes signals that AW-02 / AW-03 / AW-04 own. It must therefore degrade cleanly when those epics have not landed: §4.11 pins exactly what each block reads today and what it switches to once its owning epic ships. Home never becomes the only place a signal exists.

0. TL;DR

   /  (Home)
   ┌─────────────────────────────────────────────────────────────────────────┐
   │ Good morning, Dana.                                       Fri 6 Sep     │
   │ 3 need you · 2 working now · 7 done today · 1 failed                    │
   │ ┌─────────────────────────────────────────────────────────────────────┐ │
   │ │ Hand something to your agents…                            [ Send ]  │ │  ← composer
   │ └─────────────────────────────────────────────────────────────────────┘ │
   ├─────────────────────────────────────────────────────────────────────────┤
   │ NEEDS YOU                                              Open all (3) →   │
   │  ⚑ Approve · Publish the September notes     waiting 2h  [Yes][No]      │
   │  ? Question · Which category for these items?  waiting 20m    [Open]    │
   │  ! Escalation · Merge refused on T-241          waiting 1d     [Open]   │
   ├─────────────────────────────────────────────────────────────────────────┤
   │ TODAY AT A GLANCE                                                       │
   │   3 need you    2 working now    7 done today    1 failed today         │
   ├──────────────────────────────────┬──────────────────────────────────────┤
   │ TODAY                            │ THIS WEEK                            │
   │  ✓ 06:15  Daily inbox sweep      │  $18.42  last 7 days · Acme          │
   │    09:00  Weekly report          │  61 runs · $0.30 avg per run         │
   │    14:00  Catalog price check    │  [████████░░] 78% account-wide cap   │
   │    +2 more →                     │  Manage spend →                      │
   ├──────────────────────────────────┴──────────────────────────────────────┤
   │ WORKING NOW (2)                                        See all runs →   │
   │  ● Research agent   Reading the September changelog…            14m     │
   │  ● Writer           Drafting the September summary…              3m     │
   ├─────────────────────────────────────────────────────────────────────────┤
   │ RECENT ACTIVITY                                       Open the feed →   │
   │  07:02  Mission "Q3 catalog sweep" completed                            │
   │  06:58  Task T-240 moved to In review                                   │
   │  …                                                                      │
   ├─────────────────────────────────────────────────────────────────────────┤
   │ ▸ Your workspace   (the existing stats strip + previews, unchanged)     │
   └─────────────────────────────────────────────────────────────────────────┘

Home is the first screen after login and the most-visited route in the product. Today it answers "how big is my workspace?" — twelve total-count tiles, a list of Works, a list of Ideas. It does not answer the question the owner actually opens it with: "what happened while I was away, what needs me, and what is coming?"

This epic turns Home into a morning screen: one sentence in a composer hands out work; a decision block shows exactly what is blocked on the human and lets the easy ones be answered in place; four time-bounded counters replace guessing; today's cadence and this week's spend sit side by side; a live block names who is working and on what; and a short activity tail links into the feed. Everything that is on Home today survives, collapsed under one heading below the fold.

Three phases, each independently shippable:

P1 — The morning read. One composed summary behind the whole stack; the six read blocks; the composer creating a Task from one sentence; the existing blocks relocated below.
P2 — Answer without leaving. Inline answering in the decision block, the Home preference row (hide / reorder blocks), the composer's expand-to-full-form.
P3 — Keeping it live. 60-second background refresh with a "new since you opened this" pill, per-block retry, staleness chips, keyboard affordances.

1. Overview

A user signs in and lands on Home. A greeting names them and the date; one line under it scores the night — how many things need them, how many agents are working, how much finished, how much failed. A single text field invites one sentence, and one sentence is enough: typing "summarise every item added this week and flag the duplicates" and pressing Enter creates a Task, which lands in the Task board's Backlog lane and starts being picked up.

Below the composer, a Needs you block lists everything blocked on a human decision — an approval an agent is waiting on, a question it asked, an escalation it raised — oldest first, each row saying how long it has been waiting. Items that came with a small set of choices can be answered from Home in one click; the rest link out. Answering unblocks the work that was waiting.

Then four today counters, a Today panel showing what already ran and what is still due before midnight in the user's own timezone, a This week panel with the last seven days of spend and how far the current billing period has eaten into its cap, a Working now panel naming each running agent and the one line it last reported, and a Recent activity tail linking into the feed.

Everything the current Home shows is still there, gathered under a collapsible Your workspace heading at the bottom. A user who preferred the old screen loses nothing; a user who opens Home to find out what happened now gets an answer in one screen without scrolling into a second one.

2. Why now

2.1 The question this answers

"What happened, what needs me, and what is coming?"

This is the first question of every working day for the person Ever Works is built for: an owner who delegated work overnight and now has to decide whether anything went wrong, whether anything is stuck on them, and what is about to run. It is asked once at 07:00 and three or four more times before lunch.

2.2 What a user does today instead

To find out…	Today they must…
Whether anything is blocked on me	Read the approvals block on Home (agent action proposals only), then open Inbox for questions and escalations — two surfaces, neither of which is complete on its own
What my agents finished overnight	Open the activity page and scroll, correlating timestamps by eye; there is no "today" boundary anywhere
Whether anything failed	Read the needs-attention block, which covers errored Agents, failed generations, blocked Tasks and budget overage — but not failed Runs, which is the most common failure
Who is working right now	Open the runs list and filter it to `running`; Home shows a total Agents count and an active Agents count, neither of which is "executing right now"
What is scheduled today	Read the coming-up block, which knows only two of the seven kinds of schedule the platform actually runs and shows the next three regardless of whether they are today or next month
What I have spent this week	Open settings, then Usage & Credits, then the Costs tab, then change the window to 7 days — four navigations from Home
Hand out a new piece of work	Open the "+ New" page, pick the right one out of eleven chips (Task, Mission, Idea, Agent and the Work kinds), fill a form, submit

2.3 The five concrete gaps

No delegation from Home. The fastest path from "I want this done" to "an agent has it" is a multi-step form on another route. The most common act on an agent platform is the one with the most friction.
The decision surface is split and incomplete. Approvals render on Home; questions and escalations only render in Inbox. Neither surface says how long something has been waiting, so a decision can sit for three days without anything on screen getting louder.
Nothing on Home is time-bounded. Twelve tiles all say total. "Total Missions: 26" is the same number today, tomorrow and next month; it cannot answer "did anything happen last night?"
The two signal blocks under-report. The coming-up block silently drops five of the seven schedule kinds because it has no label for them, and it is not scoped to today. The needs-attention block has two signal kinds declared that nothing ever produces.
Spend is four clicks away and never on the morning path. The one tile that mentions money shows a month-to-date total with no cap context, no run count and no trend, and its only affordance is a deep link into settings.

2.4 Why one composed read instead of more blocks

Home currently issues roughly eighteen independent server fetches, each individually guarded so a failure degrades to an empty block. That is the right failure posture and the wrong performance posture: eighteen round trips fan out on every navigation to the most-visited route in the product, and because an empty block and a failed block look identical, a broken signal is invisible.

This epic replaces the morning stack's fan-out with one composed read that returns every block in a single response, with a per-block status so a failed block can say so instead of pretending to be empty. The inventory blocks below the fold keep their existing fetches unchanged.

3. User scenarios

3.1 Primary scenarios

S1 — The morning glance. Given a user whose agents ran overnight: 7 Runs completed, 1 failed, 2 still executing, 3 open decisions and 5 schedules due today, when they open Home at 07:04 local time, then the greeting reads Good morning, Dana. with today's date, the score line reads 3 need you · 2 working now · 7 done today · 1 failed, the Needs-you block lists 3 rows oldest first with a waiting duration on each, the Today panel shows the schedules that already fired with a tick and the ones still due with their local time, the This-week panel shows the 7-day spend total, and the Working-now panel names both running agents with the line each last reported — all of it rendered from a single server read that completes in under 800 ms at the 95th percentile.

S2 — One sentence becomes a Task. Given a user on Home, when they type summarise every item added this week and flag the duplicates into the composer and press Enter, then the field clears, a chip appears immediately under the composer reading Task created — summarise every item added this week… with a link, a Task is created with that sentence as its description and a title derived from it, the Task lands in the Task board's Backlog lane, and the chip stays visible for 60 seconds so the user can click through without hunting for it.

S3 — Answering a decision without leaving Home. Given an approval waiting for the user with two choices, Approve and Reject, when they click Approve in the Needs-you block, then the row switches to a disabled Answering… state, the answer is delivered, the row is removed with the block's count decrementing from 3 to 2, a toast reports what happened to the waiting work — Sent. The agent picked it up. when the running Run absorbed the answer, or Sent. A run resumed to answer it. when a parked Run had to be restarted — and the score line's "need you" counter updates in the same paint.

S4 — Seeing what is due before midnight. Given a user in Europe/Kyiv at 14:30 local with a daily sweep that ran at 06:15, a weekly report due at 09:00 tomorrow, and a catalog check due at 18:00 today, when they read the Today panel, then it shows exactly two rows — the 06:15 sweep marked as already run and the 18:00 catalog check as still due — and does not show tomorrow's 09:00 report, because "today" is the user's own calendar day and not a rolling window.

S5 — Spend with the cap in view. Given the active Organization Acme, in which $18.42 of usage landed in the last 7 days across 61 Runs, and an account with an account-wide monthly cap of $50 of which $39.10 has been spent so far this billing period across all of the user's Organizations, when they read the This-week panel, then the headline reads $18.42 with the sublabel last 7 days in Acme, the second line reads 61 runs · $0.30 avg per run, a neutral bar (78% is below the 80% amber threshold) reads 78% of your account-wide cap this billing period, and Manage spend → links to the costs surface with the 7-day window already selected.

S6 — Following a live run. Given two Runs executing, one started 14 minutes ago and one 3 minutes ago, when they read the Working-now panel, then the longer-running one is listed first, each row names its Agent, shows the one-line activity the Run last reported, and shows elapsed time; clicking a row opens that Run's detail.

S7 — Picking up the thread from the activity tail. Given 40 things happened since midnight, when they read the Recent-activity block, then the 8 most recent are listed newest-first with a local HH:mm time and a one-line summary, each linking to the entity it happened to, and Open the feed → opens the full feed.

S8 — Hiding a block they never use. Given a user who does not use schedules, when they open the Home block menu and toggle Today off, then the Today panel disappears, the choice persists across sessions and devices, the remaining blocks reflow to close the gap, and the block menu shows Today with an "off" state so it can be turned back on.

3.2 Edge cases, failures and races

S9 — A brand-new account. Given a user who has just signed up and has no Agents, no Missions and no Runs, when they open Home, then the score line is suppressed entirely (not rendered as 0 need you · 0 working now · 0 done today · 0 failed), the composer renders with the placeholder Hand something to your agents… and a one-line hint Describe a job in a sentence. An agent will pick it up., every read block collapses into a single empty-state card reading Nothing yet. Once your agents start working, this is where the morning report lands., and the onboarding entry point is the only call to action on the screen.

S10 — One block fails, the rest render. Given the schedule aggregation times out at 1500 ms while every other block answers, when Home renders, then the Today panel shows Couldn't load today's schedule. with a Retry button, every other block renders its real data, the page does not show a global error, the failure is recorded, and pressing Retry re-reads only that block.

S11 — Every block fails. Given the summary read itself fails (network, 500, or the whole budget exhausted), when Home renders, then the morning stack is replaced by one card reading We couldn't load your morning report. with Try again, the composer remains usable (it does not depend on the summary), and the inventory section below the fold still renders from its own fetches.

S12 — A decision is answered somewhere else first. Given the user answers the same approval from Inbox in another tab, and then clicks Approve on the stale row still rendered on Home, when the answer is delivered, then the platform reports that the item was already decided, the row is removed without an error toast, an informational toast reads Already answered elsewhere., and the block's count is recomputed from a fresh read rather than decremented locally.

S13 — The composer submission fails. Given the create call fails (offline, 500, or the create throttle of 60 per minute is exceeded), when the user pressed Enter, then the typed sentence is not lost — it stays in the field, the field regains focus, an inline message under it reads Couldn't create that Task. with a Try again button, and on a throttle response specifically the message reads You're creating these faster than we can file them. Try again in a minute.

S14 — Empty input, whitespace, and overlong input. Given the composer, when the trimmed value is shorter than 3 characters, then Send is disabled and Enter does nothing; when the value exceeds 2000 characters, then a counter appears at 1800 characters showing 1800/2000, further typing is accepted up to 2000 and refused beyond it, and Send stays enabled up to the cap.

S15 — A run that is waiting on the user appears in only one place. Given a Run whose status is running but which has raised a question and is waiting for input, when Home renders, then it appears in the Needs-you block and not in the Working-now panel, because "working" means acting, and a run waiting for a human is not acting.

S16 — A decision has been waiting too long. Given an open decision created 4 days ago, when Home renders, then its row sorts to the top of the Needs-you block, its waiting chip reads waiting 4d in the danger tone, and the block header gains a suffix reading 1 waiting over 3 days.

S17 — More decisions than the block shows. Given 14 open decisions, when Home renders, then the block shows the 5 oldest, the header reads Open all (14) →, and the score line's counter reads 14 — the preview cap never changes the count.

S18 — Timezone is unknown. Given a browser that reports no timezone, or reports one the platform does not recognise, when Home renders, then every "today" boundary falls back to UTC, and each time-bounded block carries a one-line footnote reading Times shown in UTC. so a wrong-looking count is explainable rather than mysterious.

S19 — Organization scope switches. Given a user who switches the active Organization from the workspace switcher, when Home re-renders, then every block reflects only the newly-active scope, no counter carries a number from the previous scope, and the block preference (which blocks are hidden) is per-user and therefore unchanged by the switch. The one figure that does not change is the This-week cap bar, because the spend cap is an account-wide setting; it is labelled account-wide in both scopes (FR-39a).

S20 — Background job runtime is not configured. Given an installation with no configured job runtime, so nothing will ever be dispatched, when the user creates a Task from the composer, then the Task is still created (creation does not need the runtime), the existing degraded-runtime banner is already on screen above the content, and the composer's success chip carries the suffix — nothing will run until a job runtime is configured linking to that setting.

S21 — A second tab is open. Given Home open in two tabs, when the background refresh fires in the hidden tab, then it does not fire — refresh is suspended while the document is hidden and performs one immediate read when it becomes visible again.

4. Functional requirements

Every threshold below is normative. "The summary" means the single composed read that backs the morning stack.

4.1 The screen and its order

FR-1. Home renders, in this order: greeting + score line, composer, Needs-you, Today-at-a-glance, a two-column row of Today and This-week, Working now, Recent activity, then a collapsible Your workspace section containing everything Home renders today.
FR-2. The Your workspace section is collapsed by default for an account older than 7 days and expanded by default for an account 7 days old or newer. The user's explicit expand/collapse choice overrides the default and persists (§5.4).
FR-3. Below 1024 px viewport width the two-column row stacks; below 768 px every block is full width and the score line wraps to at most two lines.
FR-4. The greeting varies by the user's local hour: Good morning (05:00–11:59), Good afternoon (12:00–17:59), Good evening (18:00–04:59), followed by the user's display name. The date is rendered in the user's locale.
FR-5. The score line renders only when at least one of its four counters is non-zero. Zero-valued counters within a rendered score line are omitted rather than shown as 0.

4.2 The composer

FR-6. The composer is a single auto-growing text field, 1 row at rest, growing to at most 6 rows, with the placeholder Hand something to your agents….
FR-7. Enter submits. Shift+Enter inserts a newline. Ctrl/Cmd+Enter also submits. Escape blurs the field without clearing it.
FR-8. Submission requires a trimmed length of at least 3 characters. The hard maximum is 2000 characters; a live counter appears at 1800.
FR-9. A successful submission creates one Task with: status = backlog (the create default, so it lands in the board's first lane), the platform's default priority, and no owner scope — it is filed against no Work, Mission, Idea, Team, Agent or Goal. description = the submitted text verbatim; title = the first sentence of the text truncated at the last word boundary at or before 80 characters, with a single trailing … when truncated. When the first sentence is shorter than 3 characters the whole text is used.
FR-10. The field is cleared and disabled while the create is in flight, and re-enabled on either outcome. A submission may not be issued while another is in flight.
FR-11. Up to 3 success chips are kept under the composer, newest first, each for 60 seconds, each linking to the created Task.
FR-12. On failure the typed text is preserved, focus returns to the field, and an inline error with a Try again action is shown. A throttle response produces a distinct message (§6.9).
FR-13. The composer's unsent text is preserved per browser and restored when Home is opened again, and is cleared on a successful submission. It is never sent anywhere until the user submits.
FR-14. The composer functions when the summary read failed (§S11) — it has no dependency on it.
FR-15. An Expand affordance opens the full Task creation form with the typed text carried into it (P2).

4.3 Needs you

FR-16. The block lists open items that are blocked on a human decision: agent questions, approvals, and escalations. Informational notices are excluded.
FR-17. Rows are ordered by age, oldest first. An item open for 72 hours or more sorts above every younger item regardless of its own age ordering within that group.
FR-18. The block previews at most 5 rows and always shows the exact total in its header link, e.g. Open all (14) →.
FR-19. Each row shows: a kind chip (Approve / Question / Escalation), the responsible Agent's name when known, the item's title truncated to 120 characters, and a waiting chip.
FR-20. The waiting chip reads waiting {n}m under 1 hour, waiting {n}h under 24 hours, and waiting {n}d at or above 24 hours. It is neutral under 24 h, amber from 24 h, and danger from 72 h.
FR-21. When the item carries between 1 and 3 choices, those choices render as inline buttons on the row and answering is possible without leaving Home. With 0 choices or more than 3, the row shows Open only.
FR-22. Answering a row optimistically removes it, then reconciles against a fresh count. An item that was already decided elsewhere is removed with an informational toast, never an error.
FR-23. When one or more items have been waiting 72 hours or more, the block header carries the suffix {n} waiting over 3 days.
FR-24. The block's empty state reads Nothing needs you right now. with a second line Your agents will raise anything they can't decide themselves.
FR-25. Failures the platform raised on its own — an errored Agent, a failed generation, a blocked Task, an exceeded budget — render below the decision rows in the same block, under the sub-heading Also broken, capped at 6 rows. They are signals, not decisions, so they are never counted in the "need you" counter.

4.4 Today at a glance

FR-26. Four counters render, each a link: need you, working now, done today, failed today.
FR-27. need you is the exact count from §4.3's decision set. working now counts Runs currently executing and not waiting on a human. done today counts Runs that finished successfully since the start of the user's local day. failed today counts Runs that ended in failure in the same window.
FR-28. failed today renders in the danger tone when greater than zero and in the neutral tone at zero.
FR-29. Counters are exact up to 999; at or above 1000 they render 999+.

4.5 Today (the day's cadence)

FR-30. The panel covers the user's local calendar day, from 00:00 to 23:59 in their timezone.
FR-31. It reads every schedule kind the platform aggregates — recurring Tasks, Agent heartbeats, Work schedules, Mission ticks, source-validation checks, data-sync polls and inbound triggers — not a subset. Each kind has its own label; no kind is dropped for lack of one.
FR-32. Rows already fired today render first, dimmed, with a tick and the local time they ran, capped at 3 with no overflow link. Rows still due render next, ascending by time, capped at 6 with a +{n} more → link.
FR-33. A row shows: local HH:mm, the owning entity's name truncated to 60 characters, the kind label, and — for a schedule that is paused or in error — its status chip. Paused and errored schedules are shown, not hidden.
FR-34. A schedule whose next fire cannot be computed is excluded from the "still due" list and does not count toward the overflow number.
FR-35. Empty state, when nothing ran and nothing is due: Nothing scheduled today. with the action Set something up →. When something ran but nothing remains: Nothing else scheduled today.
FR-36. +{n} more → opens the full schedules surface.

4.6 This week (spend)

FR-37. The headline is total AI spend over a rolling 7-day window in the active scope — the active Organization's usage when one is active, the user's personal usage otherwise (FR-69) — formatted in the account's currency, with the sublabel last 7 days in {scope}, where {scope} is the Organization's name or Personal. The window is explicitly 7 days and is never described as "this week" in copy.
FR-38. The second line reads {runs} runs · {avg} avg per run, where runs counts the active scope's Runs in the same window and avg is the scoped 7-day total divided by that count, rounded to the currency's minor unit, and reads — when the run count is zero. Spend and runs from another Organization are never mixed into either number.
FR-39. When the account has a spend cap for the current billing period, a bar shows the account's period-to-date spend as a percentage of that cap, labelled {n}% of your account-wide cap this billing period. The bar is neutral below 80%, amber from 80%, and danger from 100%.
FR-39a. The spend cap is an account-wide setting and is enforced against the account's spend in every Organization, so the cap bar and the FR-40 line cannot be scoped to one Organization. They MUST always carry the words account-wide, MUST be visually separated from the scoped headline, and MUST NOT be presented as a share of the headline figure. No other number in the panel may be account-wide.
FR-40. At or above 100% the panel adds one line: New runs are blocked. when the account blocks on overage, or Overage is allowed. when it does not.
FR-41. With no cap set the bar is replaced by the line No spend cap set. and the action Set a cap →.
FR-42. Manage spend → opens the costs surface with the 7-day window pre-selected. That surface reports account-wide spend, so the link's accessible description reads Opens account-wide spend.
FR-43. The panel is hidden entirely for a user whose account has never recorded any spend, in any scope, so a brand-new account is not shown a $0.00 meter. An account with spend elsewhere that opens an Organization with none sees $0.00 for that Organization, not a hidden panel.

4.7 Working now

FR-44. Lists Runs currently executing, excluding any Run waiting on human input (those belong to §4.3).
FR-45. Ordered longest-running first. Capped at 5 rows with a See all runs → link showing the exact total when it exceeds 5.
FR-46. A row shows: the Agent's name, the one-line activity the Run last reported truncated to 100 characters, and elapsed time formatted {n}m under 1 hour and {h}h {m}m at or above.
FR-47. A Run with no reported activity line shows Working… rather than an empty cell.
FR-48. A Run executing for 30 minutes or more carries a neutral long run chip; at 120 minutes or more the chip becomes amber and reads still going.
FR-49. Empty state: Nobody is working right now. with the action Hand out some work ↑ which focuses the composer.

4.8 Recent activity

FR-50. Lists the 8 most recent activity entries for the active scope, newest first.
FR-51. Each entry shows local HH:mm (or d MMM when older than the current local day), a one-line summary truncated to 120 characters, and links to the entity it concerns when one is resolvable.
FR-52. Open the feed → opens the activity feed.
FR-53. Empty state: Nothing has happened yet.

4.9 Freshness

FR-54. While the document is visible, the summary re-reads every 60 seconds. While hidden, no refresh is issued; one immediate read is issued on becoming visible.
FR-55. When a refresh increases the decision count, a pill appears above the Needs-you block reading {n} new since you opened this; clicking it scrolls the block into view and dismisses the pill. The pill also dismisses when the count returns to or below its value at page open.
FR-56. A refresh never moves scroll position, never steals focus, and never discards text in the composer.
FR-57. A manual refresh control is available and is also bound to the r key (§6.13).
FR-58. The server may serve a summary computed at most 10 seconds ago for the same user and scope; every response states the moment it was computed, and the UI shows updated {n}s ago next to the manual refresh control once that value exceeds 90 seconds.

4.10 Performance, limits and failure

FR-59. The summary is a single request. Its 95th-percentile server time is 800 ms; its hard ceiling is 3000 ms.
FR-60. Each block within the summary has an independent 1500 ms budget. A block that exceeds it returns a failed status and an error message key, never partial or fabricated data.
FR-61. A block reporting failure renders its own error card with Retry; retrying re-reads that block alone.
FR-62. An empty block and a failed block must be visually distinguishable. Empty says what would appear here; failed says it could not be loaded.
FR-63. Home renders its shell, greeting and composer before the summary resolves; each block renders a skeleton with the shape of its populated state.
FR-64. Home issues no more than one summary request per navigation and per refresh tick, regardless of how many blocks are visible.

4.11 Degradation against sibling epics

FR-65. Before AW-03 ships, Open all (n) → opens the Inbox. After AW-03 ships it opens My Decisions. The block itself does not change shape.
FR-66. Before AW-04 ships, Open the feed → opens the activity page. After AW-04 ships it opens the Live Feed.
FR-67. Before AW-02 ships, the composer's success chip links to the Task detail page. After AW-02 ships it links to the Task board with that Task focused.
FR-68. Home never becomes the only surface for any signal it shows. Every block links to the surface that owns its data.

4.12 Permissions, scope and safety

FR-69. Every read is scoped to the acting user, and additionally to the active Organization when one is active. There is no parameter by which a caller can name another user or another Organization.
FR-70. A resource the caller does not own is indistinguishable from one that does not exist: both answer "not found". No block leaks the existence of anything outside the caller's scope.
FR-71. Answering a decision from Home is subject to exactly the same rules, throttles and audit as answering it from its own surface. Home adds no new authority.
FR-72. The composer creates work under the same throttle the Task create path already enforces; Home does not raise it.
FR-73. Nothing in the summary contains a secret, a credential, or a raw error body from an external service. Failure messages are message keys, not provider text.
FR-74. Titles, activity lines and summaries rendered on Home are agent-authored text and are rendered as plain text, never as markup.

4.13 Accessibility and internationalisation

FR-75. Every block is a landmark region with an accessible name matching its visible heading.
FR-76. The score line is announced once on load as a single polite live region; refresh updates do not re-announce it unless a counter changed.
FR-77. The Needs-you count and the "new since you opened this" pill are polite live regions. Nothing on Home uses an assertive live region.
FR-78. Every interactive element is reachable and operable by keyboard, has a visible focus ring, and meets a contrast ratio of at least 4.5:1 in both themes.
FR-79. Every user-visible string is translatable. Counts use plural-aware messages. Times and dates are formatted in the user's locale and timezone.
FR-80. Colour is never the only carrier of meaning: every tone-carrying chip also carries text (waiting 4d, failed, paused).

5. Key entities

5.1 Already in Ever Works — read, not changed

Concept	What Home reads from it	Changed by this epic?
Task	Created by the composer; blocked Tasks feed the "Also broken" list	No — creation uses the existing path
Mission	A standing initiative, never a unit of work here: a scheduled Mission's tick is one of the seven kinds in the Today panel, and the existing Missions preview keeps its own read below the fold	No
Agent	Names in the decision rows and the working-now rows; errored Agents feed "Also broken"	No
Run	Working-now rows; the `done today` / `failed today` / `working now` counters	No — one additive index only (see plan)
Approval / Escalation / Question	The decision set behind "Needs you"	No
Schedule (the aggregated read model)	The Today panel	Widened to label all seven kinds
Activity	The recent-activity tail	No
Spend / budget	The This-week panel	No
Organization / Workspace scope	Scopes every read	No

5.2 States and transitions Home depends on

A decision item moves open → answered, and may also become archived. Home shows only open. Answering from Home performs the same transition the item's own surface performs, and the effect on the waiting work is one of:

        ┌──────────────────────────────────────────────────────────┐
        │  open                                                    │
        │    │ answered on Home / on its own surface               │
        │    ▼                                                     │
        │  answered ──┬─► the running Run absorbs the answer       │
        │             ├─► a parked Run resumes to consume it       │
        │             ├─► an approval is approved / rejected       │
        │             ├─► an escalation is resolved with the note  │
        │             └─► already decided elsewhere → no-op        │
        └──────────────────────────────────────────────────────────┘

A Run is queued → running → completed | failed | cancelled, and while running may additionally be waiting on input. Home's rule (FR-44, S15):

   running AND NOT waiting-on-input   →  Working now
   running AND waiting-on-input       →  Needs you
   completed today                    →  "done today"
   failed today                       →  "failed today"
   queued / cancelled                 →  neither

A schedule row is active | paused | disabled | error | ended. Home's Today panel shows active, paused and error (paused and errored are exactly what a morning screen must not hide); disabled and ended are excluded.

5.3 New read model — the Home summary (not a table)

One composed, read-only projection assembled per request from the entities in §5.1. It carries, per block: a status (ok / failed), the block's rows or counters, an exact total where the block previews a subset, and the moment it was computed. It is not persisted, not a user-facing noun, and introduces no vocabulary — the UI shows blocks, not "summaries".

5.4 New entity — Home preference (justified)

Home needs one small piece of durable per-user state that has nowhere to live today: which blocks a user has turned off, in what order they appear, and whether the Your workspace section is expanded.

Why it must persist server-side rather than per-browser:

Home is the landing route. A block a user has deliberately dismissed reappearing on their laptop after they hid it on their desktop is the kind of small betrayal that makes a screen feel untrustworthy.
The choice is a statement about the account's way of working (an owner with no schedules never wants the Today panel), not about one browser.
Browser storage is already used for the composer draft (FR-13), which is genuinely per-device. Mixing durable preference into the same store would make "clear site data" silently reset a deliberate choice.

Shape, conceptually: one row per user, holding the set of hidden block ids, an explicit block order, and the Your workspace expansion state. Everything about it is optional — an account with no row behaves exactly as the defaults in §4.1 and §4.2 describe, so the row is created only on first change.

It is a preference, not a new domain noun: it never appears in the product's vocabulary, is never referenced by another entity, and deleting it loses nothing but a layout choice.

5.5 Explicitly not new entities

No "briefing" or "digest" entity. The morning read is computed, never stored. The scheduled roll-up that is stored and delivered already exists and is a different feature.
No "counter" or "metric" table. Every counter is a scoped query over rows that already exist.
No new decision noun. Home reads the existing decision set; AW-03 owns its shape.
No new schedule noun. Home reads the existing aggregation.

6. UX

6.1 Where it lives

Home is the existing dashboard root route inside the existing shell (sidebar, header, chat panel, footer, degraded-runtime banner). This epic adds no route, no sidebar entry and no modal.

6.2 Populated — desktop, ≥1280 px

┌────────────────────────────────────────────────────────────────────────────────┐
│  Good morning, Dana.                                        Friday, 6 September│
│  3 need you · 2 working now · 7 done today · 1 failed            ↻ updated 4s  │
│                                                                                │
│  ┌──────────────────────────────────────────────────────────────┐  ┌────────┐  │
│  │ Hand something to your agents…                               │  │  Send  │  │
│  └──────────────────────────────────────────────────────────────┘  └────────┘  │
│   Describe a job in a sentence. An agent will pick it up.            Expand ↗  │
│   ✓ Task created — "summarise every item added this week…"  Open →             │
│                                                                                │
│ ┌────────────────────────────────────────────────────────────────────────────┐ │
│ │ NEEDS YOU  · 1 waiting over 3 days                          Open all (3) → │ │
│ ├────────────────────────────────────────────────────────────────────────────┤ │
│ │ ! Escalation  Writer   Merge refused on T-241        waiting 4d  [ Open ]  │ │
│ │ ⚑ Approve     Ops      Publish the September notes  waiting 2h  [Yes][No]  │ │
│ │ ? Question    Research Which category for these?  waiting 20m [ Open ]     │ │
│ ├────────────────────────────────────────────────────────────────────────────┤ │
│ │ ALSO BROKEN                                                                │ │
│ │ ● Agent "Scraper" errored — auto-paused after failures                 →   │ │
│ │ ● Task "Reconcile the September import" blocked                        →   │ │
│ └────────────────────────────────────────────────────────────────────────────┘ │
│                                                                                │
│ ┌────────────────────────────────────────────────────────────────────────────┐ │
│ │ TODAY AT A GLANCE                                                          │ │
│ │    3          2             7              1                               │ │
│ │  need you   working now   done today    failed today                       │ │
│ └────────────────────────────────────────────────────────────────────────────┘ │
│                                                                                │
│ ┌─────────────────────────────────────┐ ┌────────────────────────────────────┐ │
│ │ TODAY                    +2 more →  │ │ THIS WEEK           Manage spend → │ │
│ ├─────────────────────────────────────┤ ├────────────────────────────────────┤ │
│ │ ✓ 06:15  Daily sweep       heartbeat│ │  $18.42                            │ │
│ │ ✓ 07:00  Catalog sync      data sync│ │  last 7 days in Acme               │ │
│ │ ─────────────────────────────────── │ │  61 runs · $0.30 avg per run       │ │
│ │   14:00  Catalog check     recurring│ │                                    │ │
│ │   18:00  Nightly report     schedule│ │  ████████████░░░░  78%             │ │
│ │   22:00  Mission tick   ⏸ paused    │ │  of your account-wide cap          │ │
│ └─────────────────────────────────────┘ └────────────────────────────────────┘ │
│                                                                                │
│ ┌────────────────────────────────────────────────────────────────────────────┐ │
│ │ WORKING NOW (2)                                            See all runs →  │ │
│ ├────────────────────────────────────────────────────────────────────────────┤ │
│ │ ● Research   Reading the September changelog…        14m   long run        │ │
│ │ ● Writer     Drafting the September summary…          3m                   │ │
│ └────────────────────────────────────────────────────────────────────────────┘ │
│                                                                                │
│ ┌────────────────────────────────────────────────────────────────────────────┐ │
│ │ RECENT ACTIVITY                                           Open the feed →  │ │
│ ├────────────────────────────────────────────────────────────────────────────┤ │
│ │ 07:02  Mission "Q3 catalog sweep" completed                            →   │ │
│ │ 06:58  Task T-240 moved to In review                                   →   │ │
│ │ 06:41  Run finished for Agent "Research"                               →   │ │
│ │ …                                                                          │ │
│ └────────────────────────────────────────────────────────────────────────────┘ │
│                                                                                │
│  ▸ Your workspace                                                     ⋯ Blocks │
└────────────────────────────────────────────────────────────────────────────────┘

6.3 Loading

The shell, greeting and composer paint immediately. Each block paints a skeleton shaped like its populated state — never a spinner, never a collapsed zero-height region, so nothing jumps when data lands.

┌────────────────────────────────────────────────────────────────────┐
│  Good morning, Dana.                            Friday, 6 September│
│  ▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒                                  │
│  ┌──────────────────────────────────────────────┐  ┌────────┐      │
│  │ Hand something to your agents…               │  │  Send  │      │
│  └──────────────────────────────────────────────┘  └────────┘      │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ NEEDS YOU                                                      │ │
│ │ ▒▒▒▒▒▒▒▒  ▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒     ▒▒▒▒▒▒▒▒  │ │
│ │ ▒▒▒▒▒▒▒▒  ▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒     ▒▒▒▒▒▒▒▒  │ │
│ └────────────────────────────────────────────────────────────────┘ │
│ ┌────────────────────────┐ ┌───────────────────────────────────┐   │
│ │ TODAY                  │ │ THIS WEEK                         │   │
│ │ ▒▒▒▒▒▒  ▒▒▒▒▒▒▒▒▒▒▒▒   │ │ ▒▒▒▒▒▒▒▒                          │   │
│ │ ▒▒▒▒▒▒  ▒▒▒▒▒▒▒▒▒▒▒▒   │ │ ▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒                  │   │
│ └────────────────────────┘ └───────────────────────────────────┘   │
└────────────────────────────────────────────────────────────────────┘

6.4 Empty — a brand-new account

┌────────────────────────────────────────────────────────────────────┐
│  Good morning, Dana.                            Friday, 6 September│
│                                                                    │
│  ┌──────────────────────────────────────────────┐  ┌────────┐      │
│  │ Hand something to your agents…               │  │  Send  │      │
│  └──────────────────────────────────────────────┘  └────────┘      │
│   Describe a job in a sentence. An agent will pick it up.          │
│                                                                    │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │                            ☀                                   │ │
│ │              Nothing yet.                                      │ │
│ │   Once your agents start working, this is where the            │ │
│ │   morning report lands — what needs you, what ran,             │ │
│ │   and what it cost.                                            │ │
│ │                                                                │ │
│ │                 [ Set up your first agent ]                    │ │
│ └────────────────────────────────────────────────────────────────┘ │
│                                                                    │
│  ▾ Your workspace                                                  │
│    (expanded by default for accounts younger than 7 days)          │
└────────────────────────────────────────────────────────────────────┘

6.5 Empty — an established account with a quiet morning

Each block renders its own empty line rather than collapsing:

┌────────────────────────────────────────────────────────────────────┐
│ NEEDS YOU                                                          │
│   Nothing needs you right now.                                     │
│   Your agents will raise anything they can't decide themselves.    │
├──────────────────────────────┬─────────────────────────────────────┤
│ TODAY                        │ THIS WEEK          Manage spend →   │
│   Nothing scheduled today.   │   $0.00                             │
│   Set something up →         │   last 7 days in Acme               │
├──────────────────────────────┴─────────────────────────────────────┤
│ WORKING NOW                                                        │
│   Nobody is working right now.        Hand out some work ↑         │
├────────────────────────────────────────────────────────────────────┤
│ RECENT ACTIVITY                                  Open the feed →   │
│   Nothing has happened yet.                                        │
└────────────────────────────────────────────────────────────────────┘

6.6 One block failed

┌────────────────────────────────────────────────────────────────────┐
│ TODAY                                                              │
│   ⚠  Couldn't load today's schedule.                               │
│      [ Retry ]                                                     │
└────────────────────────────────────────────────────────────────────┘

6.7 The whole summary failed

┌────────────────────────────────────────────────────────────────────┐
│  Good morning, Dana.                            Friday, 6 September│
│  ┌──────────────────────────────────────────────┐  ┌────────┐      │
│  │ Hand something to your agents…               │  │  Send  │      │
│  └──────────────────────────────────────────────┘  └────────┘      │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │  ⚠  We couldn't load your morning report.                      │ │
│ │     Everything is still running — this screen just             │ │
│ │     can't see it right now.                                    │ │
│ │                        [ Try again ]                           │ │
│ └────────────────────────────────────────────────────────────────┘ │
│  ▾ Your workspace                     (renders from its own reads) │
└────────────────────────────────────────────────────────────────────┘

6.8 Over the preview cap

┌────────────────────────────────────────────────────────────────────┐
│ NEEDS YOU  · 4 waiting over 3 days                Open all (14) →  │
│ … 5 rows …                                                         │
│                        9 more waiting                              │
└────────────────────────────────────────────────────────────────────┘

6.9 Composer states

Resting
┌──────────────────────────────────────────────┐  ┌────────┐
│ Hand something to your agents…               │  │  Send  │   (disabled)
└──────────────────────────────────────────────┘  └────────┘
 Describe a job in a sentence. An agent will pick it up.

Typing (approaching the cap)
┌──────────────────────────────────────────────┐  ┌────────┐
│ summarise every item added this week…        │  │  Send  │
└──────────────────────────────────────────────┘  └────────┘
                                        1834 / 2000        Expand ↗

Sending
┌──────────────────────────────────────────────┐  ┌────────┐
│ summarise every item added this week…        │  │ ●●●    │   (disabled)
└──────────────────────────────────────────────┘  └────────┘

Sent
┌──────────────────────────────────────────────┐  ┌────────┐
│ Hand something to your agents…               │  │  Send  │
└──────────────────────────────────────────────┘  └────────┘
 ✓ Task created — "summarise every item added this week…"   Open →

Failed
┌──────────────────────────────────────────────┐  ┌────────┐
│ summarise every item added this week…        │  │  Send  │
└──────────────────────────────────────────────┘  └────────┘
 ⚠ Couldn't create that Task.   [ Try again ]

Throttled
 ⚠ You're creating these faster than we can file them. Try again in a minute.

No job runtime configured
 ✓ Task created — "summarise every item added this week…"   Open →
   Nothing will run until a job runtime is configured.  Configure →

6.10 Answering a decision inline

Before
│ ⚑ Approve   Ops   Publish the September release notes   waiting 2h  [Yes][No] │

During
│ ⚑ Approve   Ops   Publish the September release notes   Answering…            │

After (row removed, toast)
   ┌──────────────────────────────────────────────┐
   │ ✓ Sent. The agent picked it up.              │
   └──────────────────────────────────────────────┘

Already decided elsewhere
   ┌──────────────────────────────────────────────┐
   │ ℹ Already answered elsewhere.                │
   └──────────────────────────────────────────────┘

                                                          ⋯ Blocks
                                    ┌──────────────────────────────┐
                                    │ Show on Home                 │
                                    │  ☑ Needs you                 │
                                    │  ☑ Today at a glance         │
                                    │  ☐ Today                     │
                                    │  ☑ This week                 │
                                    │  ☑ Working now               │
                                    │  ☑ Recent activity           │
                                    │ ──────────────────────────── │
                                    │  Reset to defaults           │
                                    └──────────────────────────────┘

The composer and the greeting cannot be hidden — they are the screen's purpose.

6.12 Narrow viewports

< 768 px
┌───────────────────────────┐
│ Good morning, Dana.       │
│ Friday, 6 September       │
│ 3 need you · 2 working    │
│ 7 done today · 1 failed   │
│ ┌───────────────────────┐ │
│ │ Hand something to…    │ │
│ └───────────────────────┘ │
│              [   Send   ] │
│ NEEDS YOU        (3) →    │
│ …                         │
│ TODAY AT A GLANCE         │
│  3  need you              │
│  2  working now           │
│  7  done today            │
│  1  failed today          │
│ TODAY                     │
│ …                         │
│ THIS WEEK                 │
│ …                         │
└───────────────────────────┘

The four glance counters become a 2×2 grid below 768 px and a single column below 420 px. Inline decision choices stack under the row title below 768 px so a choice button is never narrower than 44 px.

6.13 Keyboard affordances

Key	Where	Does
`n`	anywhere on Home, when focus is not in a text field	Focuses the composer
`Enter`	composer	Submits
`Shift+Enter`	composer	Newline
`Ctrl/Cmd+Enter`	composer	Submits
`Escape`	composer	Blurs without clearing
`r`	anywhere on Home, when focus is not in a text field	Refreshes the summary
`Tab` / `Shift+Tab`	anywhere	Moves through blocks in visual order; each block is one tab stop group
`Enter` / `Space`	a focused decision choice	Answers
`Enter`	a focused block header link	Opens the owning surface

n and r must not collide with the shortcuts the shell already registers; they are registered through the same mechanism so the help panel lists them.

6.14 Exact user-visible copy

Where	Copy
Greeting (morning / afternoon / evening)	`Good morning, {name}.` · `Good afternoon, {name}.` · `Good evening, {name}.`
Score line	`{n} need you · {n} working now · {n} done today · {n} failed`
Freshness	`updated {n}s ago` · `updated {n}m ago`
Composer placeholder	`Hand something to your agents…`
Composer hint	`Describe a job in a sentence. An agent will pick it up.`
Composer send	`Send`
Composer expand	`Expand`
Composer counter	`{used} / 2000`
Composer success	`Task created — "{title}"` · link `Open`
Composer failure	`Couldn't create that Task.` · action `Try again`
Composer throttled	`You're creating these faster than we can file them. Try again in a minute.`
Composer runtime note	`Nothing will run until a job runtime is configured.` · link `Configure`
Needs-you heading	`Needs you`
Needs-you overdue suffix	`{n} waiting over 3 days`
Needs-you link	`Open all ({n})`
Needs-you overflow footer	`{n} more waiting`
Kind chips	`Approve` · `Question` · `Escalation`
Waiting chip	`waiting {n}m` · `waiting {n}h` · `waiting {n}d`
Row action	`Open`
Answering state	`Answering…`
Answer success (steered)	`Sent. The agent picked it up.`
Answer success (resumed)	`Sent. A run resumed to answer it.`
Answer success (approval)	`Approved.` · `Rejected.`
Answer success (escalation)	`Resolved.`
Answer already decided	`Already answered elsewhere.`
Answer failure	`Couldn't send that answer.` · action `Try again`
Needs-you empty	`Nothing needs you right now.` / `Your agents will raise anything they can't decide themselves.`
Also-broken heading	`Also broken`
Glance heading	`Today at a glance`
Glance labels	`need you` · `working now` · `done today` · `failed today`
Today heading	`Today`
Today kind labels	`recurring task` · `heartbeat` · `work schedule` · `mission tick` · `source check` · `data sync` · `trigger`
Today ran marker	`ran at {time}`
Today overflow	`+{n} more`
Today empty	`Nothing scheduled today.` · action `Set something up`
Today nothing left	`Nothing else scheduled today.`
Today paused chip	`paused` · `error`
This-week heading	`This week`
This-week sublabel	`last 7 days in {scope}` (Organization name, or `Personal`)
This-week second line	`{n} runs · {amount} avg per run`
This-week cap bar	`{n}% of your account-wide cap this billing period`
This-week cap note	`The cap applies across all your Organizations.`
This-week link description	`Opens account-wide spend`
This-week blocked	`New runs are blocked.`
This-week overage	`Overage is allowed.`
This-week no cap	`No spend cap set.` · action `Set a cap`
This-week link	`Manage spend`
Working-now heading	`Working now ({n})`
Working-now fallback line	`Working…`
Working-now chips	`long run` · `still going`
Working-now link	`See all runs`
Working-now empty	`Nobody is working right now.` · action `Hand out some work`
Recent-activity heading	`Recent activity`
Recent-activity link	`Open the feed`
Recent-activity empty	`Nothing has happened yet.`
Block error	`Couldn't load {block}.` · action `Retry`
Whole-summary error	`We couldn't load your morning report.` / `Everything is still running — this screen just can't see it right now.` · action `Try again`
New-since pill	`{n} new since you opened this`
Timezone footnote	`Times shown in UTC.`
Workspace section	`Your workspace`
Block menu	`Blocks` / `Show on Home` / `Reset to defaults`
First-run empty	`Nothing yet.` / `Once your agents start working, this is where the morning report lands — what needs you, what ran, what it cost.` · action `Set up your first agent`

7. Out of scope

Owning the decision queue. Home previews decisions; AW-03 owns their shape, filters, bulk actions and the dedicated surface.
Owning the feed. Home shows an 8-row tail; AW-04 owns the feed, its grouping and the "while you were away" summary.
Owning the Task board. The composer files a Task; AW-02 owns lanes, cards, staleness and steering.
Owning cost analysis. Home shows one 7-day number and one cap bar; AW-17 owns caps, credits, per-agent and per-model breakdowns.
Owning the calendar. Home shows today; AW-10 owns the calendar, heartbeats and never-runs.
Routing the composer's sentence to a specific Agent. The composer creates a Task with no assignee; which Agent picks it up is the Task board's and the assignment path's concern, not Home's.
Rich composition. No attachments, no @ mentions, no # references and no markdown in the composer in this epic.
A push transport. Freshness is polling on a 60-second cadence. Replacing polling with a push channel is a platform-wide change, not a Home change.
Multi-user or shared Home. Home is the acting user's own screen; shared read-only dashboards are AW-18.
Reordering blocks by drag. P2 stores an order; the only editor in this epic is show/hide plus reset. Drag ordering is a follow-up.
Removing or rewriting the existing Home blocks. They move; they do not change.
A "good morning" email or notification. The scheduled roll-up that is delivered outside the product already exists and is a different feature.

8. Acceptance criteria

A reviewer can run this list against a build.

Layout and shell

Home renders the blocks in the order given in FR-1.
The Your workspace section contains every block Home renders today, and each of those blocks still works (links, actions, empty states).
Your workspace is collapsed by default on an account older than 7 days and expanded on a newer one; toggling it persists across a sign-out and back in.
At 1023 px the Today / This-week row stacks; at 767 px every block is full width and the glance counters form a 2×2 grid.
The greeting changes across the 05:00 / 12:00 / 18:00 local boundaries.
With all four counters at zero, the score line is absent — not four zeros.

Composer

Needs you

With 1 question, 1 approval and 1 escalation open, all three appear, and an informational notice does not.
With 14 open items, 5 render, the header reads Open all (14) →, the footer reads 9 more waiting, and the score line reads 14.
Rows are oldest first; an item open 4 days sorts above one open 2 days even when the newer one was created by a different mechanism.
An item open 23 h 59 m shows a neutral chip; at 24 h it is amber; at 72 h it is danger and the header suffix appears.
An item with 2 choices renders 2 inline buttons; one with 5 choices renders Open only.
Clicking a choice removes the row, decrements the count, and toasts the routed outcome.
Answering the same item elsewhere first produces the informational "already answered" toast and no error.
Errored Agents, failed generations, blocked Tasks and exceeded budgets appear under Also broken, capped at 6, and are not counted in need you.
Every rendered title is plain text: an item titled with markup renders the markup as characters.

Today at a glance

The four counters match the blocks they summarise on the same paint.
A Run that is running but waiting on input is counted in need you, not working now, and does not appear in the Working-now block.
A Run that completed at 23:59 yesterday local is not counted in done today; one that completed at 00:01 today is.
failed today is danger-toned above zero and neutral at zero.
A count of 1200 renders 999+.

Today

All seven schedule kinds render with their own label; none is dropped.
A schedule due tomorrow does not appear; one due at 23:55 today does.
A schedule that fired at 06:15 today renders dimmed, above the due list, with ran at 06:15.
Paused and errored schedules render with their status chip.
A schedule with no computable next fire is absent and is not counted in the overflow number.
With 9 still due, 6 render and the link reads +3 more →.
With nothing at all, the empty state and its Set something up → action render.

This week

The headline equals the sum of the active scope's usage over the same 7-day window, to the cent; with all of an account's usage in one Organization it equals the costs surface's 7-day total.
With usage in two Organizations, each Organization's headline, run count and average count only that Organization's usage and Runs, while the cap bar shows the same percentage in both and reads account-wide.
The average line reads — when the 7-day run count is zero.
With a cap and 78% used, the bar is neutral; at 80% amber; at 100% danger with the blocked-or-overage line.
With no cap, No spend cap set. and Set a cap → render instead of a bar.
Manage spend → lands on the costs surface with the 7-day window selected.
An account with no recorded spend does not render the panel at all.

Working now

Two running Runs render, longest first, with agent name, activity line and elapsed time.
A Run with no activity line shows Working….
A Run at 31 minutes carries long run; at 121 minutes it carries still going in amber.
With 7 running, 5 render and See all runs → shows the exact total.
With none, the empty state renders and its action focuses the composer.

Recent activity

Exactly 8 rows render when at least 8 exist, newest first.
Rows from today show HH:mm; older rows show d MMM.
Each row links to the entity it concerns when one is resolvable, and is non-interactive when none is.

Freshness and failure

With the tab visible the summary re-reads once per 60 s; hiding the tab stops it; showing it re-reads immediately.
A refresh that raises the decision count shows the pill; clicking it scrolls to the block and dismisses it.
A refresh mid-typing does not clear the composer or move focus or scroll.
Forcing one block to exceed 1500 ms renders that block's error card with Retry, leaves the others populated, and Retry re-reads only that block.
Forcing the whole read to fail renders the whole-summary error, keeps the composer working, and leaves the Your workspace section populated.
Home issues exactly one summary request per navigation and one per tick.
Empty and failed states are visually and textually distinct for every block.

Switching the active Organization changes every scoped number on the screen and leaves no stale value; the only unchanged figure is the cap bar labelled account-wide.
No request Home issues accepts a user id or an organization id parameter.
An id belonging to another account answers "not found" identically to an id that does not exist.
No response behind Home contains a credential, a token or a raw external error body.
Every block is a named landmark; a screen reader announces the score line once on load.
The whole screen is operable by keyboard alone, including answering a decision.
Every chip that carries a tone also carries text.
Every string on the screen resolves from a translation key; switching locale leaves no English behind, and no key contains a literal . in its leaf name.
With the browser timezone unavailable, counts use UTC and the footnote renders.

9. Open questions

[NEEDS CLARIFICATION: is "this week" a rolling 7 days or the calendar week?] This spec pins a rolling 7-day window because that is the window the costs surface already supports, and it labels it last 7 days so the copy is honest. If Product wants a Monday-to-now calendar week, the costs surface must first accept a window that is not one of its three fixed lengths.
[NEEDS CLARIFICATION: where does the user's timezone come from long-term?] This spec takes it from the browser and validates it server-side, falling back to the account's notification timezone and then to UTC. A first-class account timezone would be better for every time-bounded surface (this one, the digest, quiet hours) and is worth its own small epic.
[NEEDS CLARIFICATION: should "done today" count Runs or Tasks?] This spec counts Runs, because a Run is the unit that finishes. An owner may read "done" as "Tasks moved to Done". If so, the counter should show both, which costs a column of width the glance row does not have.
[NEEDS CLARIFICATION: should the composer be able to address an Agent?] Out of scope here. If a later epic adds @name addressing, Home should adopt it rather than inventing its own syntax.
[NEEDS CLARIFICATION: does the "Also broken" list belong in Needs you?] It is currently a sub-list so the morning screen has one place to look. It could equally be its own block. The counter treatment (never counted as "need you") is the part that matters and should survive either choice.
[NEEDS CLARIFICATION: what happens to the existing mock activity component?] A component named for recent activity exists in the codebase, is not rendered anywhere, contains hard-coded sample rows and hard-coded English. This epic neither renders nor deletes it. Deleting it needs an explicit decision.
[NEEDS CLARIFICATION: should hidden blocks be per-Organization?] This spec makes the preference per-user, so it survives an Organization switch. An owner running two very different Organizations may want per-scope layouts.
[NEEDS CLARIFICATION: does the 60-second cadence need to adapt?] A workspace with nothing running does not need a read a minute. Backing off to 5 minutes after 10 consecutive unchanged reads would cut load substantially; it is deliberately left out of P3 until we have real numbers.

10. Non-functional requirements

NFR-1. Summary p95 ≤ 800 ms server time, hard ceiling 3000 ms, per-block budget 1500 ms.
NFR-2. Home issues one summary request per navigation and per refresh tick; the total request count for the morning stack is 1, not 18.
NFR-3. The summary response is cacheable per user and scope for 10 seconds and must never be cached by a shared cache or a browser disk cache.
NFR-4. Every query behind the summary is bounded: no unbounded scan, no query without a row limit, and no per-row follow-up query.
NFR-5. Home's added client bundle weight is ≤ 40 KB gzipped over today's.
NFR-6. Cumulative layout shift attributable to blocks resolving is ≤ 0.1; skeletons match populated heights within 8 px.
NFR-7. A failing block never fails the page, and a failing page never fails the shell.
NFR-8. Every failure is recorded with the block that failed and the reason; a silently empty block is a defect, not a degradation.

11. Constitution gates

Gate	Status	Note
I — Plugin-first	✅	No external integration. Home reads platform data only.
II — Capability-driven	✅	No plugin id appears anywhere in this epic.
III — Source-of-truth repos	✅	No content is read from or written to a repository.
IV — Job runtime	✅	Home dispatches no background work. The composer creates a Task through the existing path; whatever that path dispatches is unchanged.
V — Forward-only migrations	✅	One additive preference table and one additive index, both forward-only, both shipping with the change that needs them.
VI — Tests first-class	✅	Unit tests for every threshold in §4, controller tests for the new read, an end-to-end test per user scenario in §3.1 and for §3.2's failure paths.
VII — Secrets	✅	Nothing Home reads is a secret; failure copy is keyed, never provider text (FR-73).
VIII — Plugin counts	✅	No plugin list changes.
IX — Behaviour-first spec	✅	This document names no class, path or library.
X — Backwards compatibility	✅	No existing endpoint, response shape or i18n key changes meaning. One new read is added; existing reads keep working.

12. Cross-references

Program overview and rules — ../README.md
Task board (the lane the composer's Task lands in) — ../AW-02-task-board/spec.md
My Decisions (owns the decision queue Home previews) — ../AW-03-decision-queue/
Live Feed (owns the feed Home tails) — ../AW-04-live-feed/
Runs & receipts (owns the Runs Home counts) — ../AW-09-runs-receipts/
Schedules & calendar (owns the cadence Home shows for today) — ../AW-10-schedules-calendar/
Costs & caps (owns the meters Home summarises) — ../AW-17-costs-caps/
Command palette (registers Home's keys in the shared shortcut surface) — ../AW-01-command-palette/spec.md
Existing schedules aggregation this epic widens — ../../schedules/spec.md
Implementation plan — ./plan.md
Task breakdown — ./tasks.md

0. TL;DR​

1. Overview​

2. Why now​

2.1 The question this answers​

2.2 What a user does today instead​

2.3 The five concrete gaps​

2.4 Why one composed read instead of more blocks​

3. User scenarios​

3.1 Primary scenarios​

3.2 Edge cases, failures and races​

4. Functional requirements​

4.1 The screen and its order​

4.2 The composer​

4.3 Needs you​

4.4 Today at a glance​

4.5 Today (the day's cadence)​

4.6 This week (spend)​

4.7 Working now​

4.8 Recent activity​

4.9 Freshness​

4.10 Performance, limits and failure​

4.11 Degradation against sibling epics​

4.12 Permissions, scope and safety​

4.13 Accessibility and internationalisation​

5. Key entities​

5.1 Already in Ever Works — read, not changed​

5.2 States and transitions Home depends on​

5.3 New read model — the Home summary (not a table)​

5.4 New entity — Home preference (justified)​

5.5 Explicitly not new entities​

6. UX​

6.1 Where it lives​

6.2 Populated — desktop, ≥1280 px​

6.3 Loading​

6.4 Empty — a brand-new account​

6.5 Empty — an established account with a quiet morning​

6.6 One block failed​

6.7 The whole summary failed​

6.8 Over the preview cap​

6.9 Composer states​

6.10 Answering a decision inline​

6.11 The block menu (P2)​

6.12 Narrow viewports​

6.13 Keyboard affordances​

6.14 Exact user-visible copy​

7. Out of scope​

8. Acceptance criteria​

Layout and shell​

Composer​

Needs you​

Today at a glance​

Today​

This week​

Working now​

Recent activity​

Freshness and failure​

Scope, security and a11y​

9. Open questions​

10. Non-functional requirements​

11. Constitution gates​

12. Cross-references​