## Summary
PR #6732 taught the ACP harness to honor `BUZZ_ACP_SESSION_POLICY`, but
the desktop exposed that policy as one global experiment. Turning it on
changed every managed agent at once, even though conversation scope is
part of how an individual agent should behave. Some agents need
continuity across a channel, while agents such as validators may need
clean context for each thread.
This moves the choice onto the agent definition under Advanced as
**Conversation context**, with **Entire channel** and **Each thread**
options. **Entire channel** remains the default, so existing definitions
keep their current behavior.
The policy now travels through create and update IPC, returned agent
summaries, persona events, snapshots, teams, and community catalog
copies. Launches pass the effective definition value to the harness.
Editing a running agent uses the same lifecycle as a model edit: it
shows the restart-required state, leaves the deployed process untouched,
and applies the new policy on restart.
The global Thread Scoped ACP Sessions experiment and its desktop
command/state have been removed because the setting no longer has a
useful global meaning. There is intentionally no migration for the old
preview override. Users who enabled it can choose **Each thread** for
the agents that need it.
Unknown or null policy values degrade to **Entire channel** across
stored agents, persona events, and catalog projections. This keeps
forward-version data from dropping complete agent records.
### Related issue
No matching issue found. Builds on #6732.
### Testing
- Focused Playwright create flow selected **Each thread** and verified
`create_persona` received `sessionPolicy: "thread"`.
- Focused Playwright catalog flow added a thread-scoped agent and
verified the local copy stored `session_policy: "thread"`.
Generated with Codex
---------
Signed-off-by: Salman Mohammed <smohammed@squareup.com>
## Why
Thread-scoped ACP sessions from #6732 need an opt-in desktop rollout
that preserves today’s channel policy by default.
## What
- Add the default-off “Thread Scoped ACP Sessions” toggle under Settings
→ Experiments using the existing feature-override persistence.
- Map the persisted setting to `BUZZ_ACP_SESSION_POLICY=channel|thread`
for every local and provider-backed managed ACP launch. Changes apply
when managed agents next start; DMs remain conversation-scoped by the
backend.
- Keep the desktop setting authoritative over descriptor environment and
cover the UI, persistence, local spawn, provider payload, and existing
Projects/Workflows experiments.
## Risk Assessment
Low. The experiment defaults off and explicitly preserves `channel`;
changes are limited to managed-agent launch configuration. Existing
running agents are unchanged until their next start.
## Testing
- `. ./bin/activate-hermit && just ci` — passed on the final restacked
tree, including 5,674 desktop tests, 2,796 Tauri tests, and 1,860 mobile
tests.
- `cd desktop && pnpm build:e2e && pnpm exec playwright test
tests/e2e/experimental-features.spec.ts --project=smoke` — 1 passed.
- `. ./bin/activate-hermit && cargo test --manifest-path
desktop/src-tauri/Cargo.toml session_policy --lib` — 4 passed.
- `. ./bin/activate-hermit && cargo test --manifest-path
desktop/src-tauri/Cargo.toml commands::agents::deploy::tests --lib` — 15
passed.
- `. ./bin/activate-hermit && cargo test -p buzz-backend-kubernetes
--test wire_fixtures` — 4 passed.
## Stack Info
Stacked on #6732. This PR depends on #6732 and should merge after it.
Generated with Codex
---------
Signed-off-by: Salman Mohammed <smohammed@squareup.com>
Signed-off-by: Leo <5faf251baee50ee6bcde338aef6acdd70bb3e60115664c2cd490d94a55dfc488@buzz.block.builderlab.xyz>
Co-authored-by: Leo <5faf251baee50ee6bcde338aef6acdd70bb3e60115664c2cd490d94a55dfc488@buzz.block.builderlab.xyz>
## Why
The Inbox mixed overlapping feed categories with personal work queues,
so **All** was not actually comprehensive and several filters did not
make it clear why an item appeared. Threads and DMs could produce one
row per event instead of one row per conversation, drafts were hidden
until selected, and reminders appeared through multiple competing
presentations.
This refactor makes the **Inbox** a focused, conversation-oriented place
to catch up on work relevant to you. It is intentionally not a mirror of
every unread event in every channel.
## What changed
- Keep the destination named **Inbox** and use the standard Lucide bell
icon.
- Refocus **All** on DMs, mentions, thread replies, needs-action items,
replies from agents the user owns or controls, due reminders, and active
drafts.
- Exclude generic top-level channel traffic and updates from agents the
user does not own or control.
- Group each thread or DM into one row, sorted by latest activity.
- Resume an unread conversation at its oldest unread message while
opening the full thread or DM in the detail pane.
- Reuse the existing **New** divider at the unread boundary.
- Make the detail title a direct link to the canonical conversation.
- Give Reminders and Drafts the same list/detail interaction and
location metadata as conversation rows.
- Separate Reminders and Drafts from message filters with a subtle
divider, without adding another labeled section.
- Put reminder and draft counts beside their corresponding filter labels
instead of on the generic filter button.
- Preserve the selected conversation when switching filters if it
remains valid; otherwise select a valid replacement without flashing
stale detail.
- Use filter-specific empty states and rename the options toggle to
**Show unread only**.
- Ship the focused behavior directly. The earlier experiment gate,
Custom view, and default-view controls have been removed from this PR to
keep the first pass focused.
## Filter model
| Filter | What appears |
| --- | --- |
| **All** | One row per personally relevant conversation, plus due
reminders and active drafts. Includes DMs, mentions, thread replies,
explicit needs-action items, and replies from agents the current user
owns or controls. Excludes generic top-level channel traffic, other
agents' updates, and reminders that are not due yet. |
| **Mentions** | Conversations containing a direct mention. Each
conversation appears once and opens with full context. |
| **Threads** | Conventional threaded replies, grouped to one row per
thread. Broadcast replies are not treated as conventional thread
replies. |
| **Needs action** | Feed items explicitly classified as requiring
action. |
| **Agents** | Conversations whose representative response was authored
by an agent the current user owns or controls, including top-level DM
responses. If a human replies afterward, the conversation leaves this
filter until an owned agent responds again. |
| **Reminders** | All pending reminders, including upcoming reminders
that stay out of **All** until they are due. |
| **Drafts** | Active drafts, ordered by their last real edit time. |
## Grouping, ordering, and state
- A thread or DM creates one Inbox row rather than one row per event.
- An unread conversation resumes at its oldest unread message so
intervening context is not skipped.
- Conversation rows still sort by their latest activity.
- The detail pane opens the full available conversation and shows the
shared **New** divider before the first unread message.
- Upcoming reminders appear only in **Reminders**.
- When a reminder becomes due, it enters **All** at its trigger time. If
its source conversation is already represented, the reminder state
merges into that row instead of creating a duplicate; otherwise it
appears as a standalone reminder row.
- A due reminder can enrich a row in another relative filter when that
conversation already qualifies for the filter. Reminder lifecycle
remains separate from message read state.
- Drafts appear in **All** by their last real edit time. Opening an
unchanged draft does not move it to the top.
- Reminder and draft rows show their location as `In #channel` or `In DM
with <name>`.
- **Show unread only** hides reminder and draft work queues because they
do not share message unread semantics.
## Removed or narrowed
- **Remove the old Activity filter.** It overlapped with All while still
omitting items All now includes.
- **Narrow Agents.** It no longer gathers every agent participating in a
shared thread or subsequent human follow-ups.
- **Remove duplicate reminder presentations.** The aggregate
pending-reminders jump and duplicate generic feed rows are replaced by
one list/detail model.
- **Remove Custom and default-view settings from this pass.** They added
considerable state and UI before the core model had been validated.
- **Do not add section labels for Reminders and Drafts.** A divider
communicates the distinction without creating another hierarchy in the
menu.
## Risk assessment
Medium implementation risk because this changes composition, grouping,
ordering, read behavior, and personal queues in a primary desktop view.
The implementation is scoped to the desktop UI and its local feed
projection; it does not change relay schemas or public APIs.
## Testing
- Desktop formatting, lint, file-size, text-size, and TypeScript checks
passed.
- Desktop unit suite: **3,663 passed, 0 failed**.
- Desktop E2E production build passed.
- Playwright smoke coverage across every spec touching this surface
(`channels`, `smoke`, `profile`, `project-inbox`, `community-rail`,
`integration`, `drafts-screenshots`): **118 passed, 0 failed**.
- Full Playwright smoke project: **732 passed, 1 skipped**. Three local
failures were investigated and cleared — `community-rail` keyboard
reorder passed on re-run (flaky), while `relay-reconnect:97` and
`video-attachment:223` are untouched by this commit (the only change to
shared `tests/helpers/bridge.ts` is a comment) and pass in CI.
- Unit coverage includes focused All matching, owned-agent filtering,
conversation grouping, oldest-unread selection, selection stability,
chronological reminder/draft composition, trigger-time reminder
ordering, and duplicate reminder suppression.
## Update: July 27, 2026
The naming decision is settled: the surface stays **Inbox**. An earlier
pass in this branch had renamed it to **Activity**; that rename has been
reverted in `9c00d2d6e`, which is naming-only and changes no behavior.
The revert covers file names, component/hook/type/constant identifiers,
the sidebar label and tooltip, the `Inbox options` and `Filter inbox:`
aria-labels, and the corresponding test names, test ids, and fixture
ids.
Three things were deliberately left as `activity`:
- **The feed API contract** — the `activity` / `agent_activity`
categories, the `feed.activity` and `feed.agentActivity` keys, and the
`types=` query parameter. These are the server's names, not the
surface's.
- **Plain-noun usage** — empty states such as "No activity yet", plus
`latestActivityAt` and `PROJECT_ACTIVITY_KINDS`.
- **Pre-existing agent, project, and profile activity code**, which
refers to a different concept entirely.
The earlier experiment-gate approach has also been dropped, so
`tests/helpers/bridge.ts` no longer claims that an Activity preview
feature exists — `preview-features.json` has no such entry and the seed
helper enables every desktop feature.
Generated with Codex
---------
Signed-off-by: Clay Delk <clay.delk@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
