Skip to content

feat(agents): group sequential activity tools#1571

Merged
tellaho merged 7 commits into
mainfrom
tho/activity-row-grouping
Jul 7, 2026
Merged

feat(agents): group sequential activity tools#1571
tellaho merged 7 commits into
mainfrom
tho/activity-row-grouping

Conversation

@tellaho

@tellaho tellaho commented Jul 7, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator

Category: improvement
User Impact: Agent activity feeds are quieter because routine consecutive tool work collapses into clearer summary rows with animated count updates as the work grows.
Problem: Consecutive tool calls made agent transcripts feel busy, especially when read/search/shell work alternated and defeated same-kind grouping. A first pass at broader grouping also exposed redundant nested summaries like Ran 16 tool calls expanding to Ran 12 commands.
Solution: This PR keeps user prompts, messages, failures, permissions, lifecycle/status, and metadata visible while broadening only eligible routine tool bursts into one mixed summary. Mixed summaries now expand directly to leaf tool rows, and count-led summary labels reuse the existing local AnimatedCount flourish instead of vendoring number-flow or adding a dependency.

File changes

desktop/playwright.config.ts
Removes obsolete screenshot/test wiring that no longer applies to the updated activity surface.

desktop/scripts/check-file-sizes.mjs
Drops a temporary file-size allowance that is no longer needed after the activity-scope cleanup.

desktop/src-tauri/src/managed_agents/personas.rs
Keeps persona test fixtures aligned with the current managed-agent activity assumptions.

desktop/src-tauri/src/managed_agents/personas/tests.rs
Updates managed-agent persona tests after removing observer/working-signal plumbing.

desktop/src/app/AppShell.tsx
Removes old app-level active-agent wiring now handled through the streamlined activity surfaces.

desktop/src/app/navigation/useAppNavigation.ts
Keeps navigation behavior aligned with the simplified activity/profile routing.

desktop/src/features/agents/activeAgentTurnsStore.test.mjs
Removes tests for the deleted active-agent turn store.

desktop/src/features/agents/activeAgentTurnsStore.ts
Simplifies active-turn tracking after removing the old activity signal store.

desktop/src/features/agents/agentWorkingSignal.test.mjs
Removes tests for the deleted agent working-signal helper.

desktop/src/features/agents/agentWorkingSignal.ts
Removes the old derived working-signal path in favor of the current transcript/session activity model.

desktop/src/features/agents/ui/AgentSessionTranscriptList.tsx
Renders turn-grouped transcript segments, mixed tool-run summaries, flattened summary expansion content, and animated count-led summary labels.

desktop/src/features/agents/ui/ManagedAgentRow.tsx
Keeps managed-agent row activity display aligned with the updated activity model.

desktop/src/features/agents/ui/activityRenderClasses/ActivityRow.tsx
Adds support for splitting count-led summary objects so the numeric prefix can be rendered with AnimatedCount while preserving the readable label.

desktop/src/features/agents/ui/activityRenderClasses/activityRowLabelParts.test.mjs
Adds coverage for count-led label splitting and non-count label fallbacks.

desktop/src/features/agents/ui/agentSessionTranscript.test.mjs
Updates transcript expectations for status/lifecycle handling in the quieter activity feed.

desktop/src/features/agents/ui/agentSessionTranscript.ts
Keeps transcript item classification compatible with the new grouping and status guardrails.

desktop/src/features/agents/ui/agentSessionTranscriptGrouping.test.mjs
Adds coverage for same-kind grouping, mixed consecutive tool bursts, guardrails for failures/status/suppressed rows, steer context placement, flattened mixed-summary expansion, and burst boundaries.

desktop/src/features/agents/ui/agentSessionTranscriptGrouping.ts
Builds display blocks by turn, routes prompt/context metadata into the right UI affordance, groups eligible same-kind tools, groups broader mixed tool bursts, and prevents nested duplicate summaries.

desktop/src/features/agents/ui/useManagedAgentActions.ts
Keeps managed-agent action state compatible with the simplified activity signal flow.

desktop/src/features/agents/useAgentObserverIngestion.test.mjs
Removes tests for deleted observer-ingestion plumbing.

desktop/src/features/agents/useAgentObserverIngestion.ts
Removes old observer-ingestion state that is no longer needed for activity display.

desktop/src/features/agents/useOpenAgentActivity.test.mjs
Removes tests for deleted open-agent activity plumbing.

desktop/src/features/agents/useOpenAgentActivity.ts
Removes old open-activity state in favor of current profile/thread-panel activity entry points.

desktop/src/features/agents/usePreventSleep.ts
Keeps prevent-sleep behavior aligned with the updated agent activity lifecycle.

desktop/src/features/channels/ui/AgentSessionThreadPanel.tsx
Adds breathing room to the activity transcript panel and keeps the thread panel aligned with the new display blocks.

desktop/src/features/channels/ui/BotActivityBar.tsx
Simplifies bot activity display now that transcript grouping carries more of the feed-noise reduction.

desktop/src/features/channels/ui/ChannelPane.tsx
Updates channel pane activity wiring to match the simplified agent activity model.

desktop/src/features/channels/ui/ChannelScreen.tsx
Keeps channel screen state aligned with the updated activity surfaces.

desktop/src/features/channels/ui/useChannelActivityTyping.test.mjs
Removes tests for deleted channel activity typing helper.

desktop/src/features/channels/ui/useChannelActivityTyping.ts
Removes old typing/activity helper that is no longer part of the current activity path.

desktop/src/features/profile/ui/UserProfilePanel.tsx
Keeps profile-panel activity entry points aligned with the updated agent session transcript panel.

desktop/src/features/profile/ui/UserProfilePanelSections.tsx
Updates profile section wiring for the simplified agent activity model.

desktop/src/features/profile/ui/UserProfilePopover.tsx
Keeps profile popover activity affordances aligned with the current activity flow.

desktop/src/features/sidebar/lib/useActiveWorkingChannelsById.ts
Simplifies active working-channel derivation after removing old agent working signals.

desktop/src/features/workspaces/useWorkspaceInit.ts
Removes initialization wiring for deleted activity/working-signal state.

desktop/src/testing/e2eBridge.ts
Updates the test bridge to support the current activity transcript scenarios.

desktop/tests/e2e/activity-scope-label-screenshots.spec.ts
Removes stale screenshot coverage for the previous activity-scope label behavior.

Reproduction Steps

  1. Open an agent activity transcript with a turn containing consecutive routine tool calls.
  2. Confirm same-kind runs still collapse into specific summaries such as Read 3 files or Ran 12 commands when they stand alone.
  3. Confirm mixed/interleaved routine bursts collapse into one summary such as Ran 16 tool calls.
  4. Expand that mixed summary and confirm it shows the underlying leaf tool rows directly, not a nested duplicate summary like Ran 12 commands.
  5. Let a live grouped tool burst grow and confirm the numeric count rolls with the existing odometer-style animation while reduced-motion users get static text.
  6. Confirm prompts, assistant/user messages, errors, permissions, status/lifecycle rows, and metadata boundaries remain visible instead of being swallowed by grouping.

Screenshots/Demos

Marge captured the latest build at 2e2a99b5 with a more realistic research/edit/verify turn. The screenshots show the intended two-tier drill-down: collapsed group, opened group with individual tool rows still closed, then one leaf row opened for detail.

State Preview What to look for
Collapsed image The whole research/edit/verify loop reads as one grouped summary row.
Group open, leaves closed image Opening the group reveals the tool-call sequence as collapsed leaf rows, without opening every result at once.
One leaf open image Each leaf can be expanded independently to inspect params and result details.

Regression proof — the earlier screenshot pair confirmed the nested-summary bug is gone: before the fix a mixed group expanded into a redundant nested Ran 12 commands row; after the fix it expanded directly to leaf tool rows.

Before After
image image

Validation

  • Marge reproduced the nested-summary bug at parent 85b244b7 and verified the fix at 9601b73a with before/after screenshots.
  • biome check . clean.
  • tsc --noEmit clean.
  • Full desktop unit suite passed at 1902/1902 after the AnimatedCount update.
  • Push hooks passed on 2e2a99b5.

Coordination: buzz://message?channel=2393ad5d-dbb4-4ea3-9b3b-36f432443238&thread=fc0758f5bcf34658ad3ed6bb13eb6fbbb7f65e3b48128381a774c57429b95ec0

npub1223z34hd7vtwc6qj4s7flsxkj644nlre2nthu7lrrmkumhu3xddsrx9r6w and others added 7 commits July 6, 2026 21:06
@tellaho
feat(desktop): group mixed consecutive tool runs in activity transcript
5aa6856 
Add a fallback grouping pass that collapses adjacent eligible tool rows
of differing kinds into one 'Ran N tool calls' summary when the existing
same-kind pass leaves them expanded. Same-kind summaries keep their
specific labels and win over mixed grouping. Messages, permissions,
errors (including failed tools), and lifecycle/status rows break runs so
intervention points stay visible. Mixed summaries preserve child order,
expand to the original rows, and aggregate diff stats when edits are in
the run.

Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
@tellaho
fix(desktop): keep failed/status/suppressed tool rows out of same-kin…
24a01e3 
…d grouping

Unify grouping eligibility across both passes: the same-kind pass now
uses the same isGroupingEligible gate as the mixed pass, so failed tools
(isError or renderClass error), status-like hook rows, and suppressed
rows can no longer collapse into a same-kind summary and always break
runs. Adds tests proving failed/status/suppressed runs stay expanded and
that an ineligible row splits a same-kind run.

Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
@tellaho
fix(desktop): route steer prompt context into the checks-icon ingress…
78a68da 
… dialog

Steer prompts (_goose/unstable/session/steer) upserted their prompt
context metadata without an acpSource, so display grouping never folded
it into the prompt bundle and it leaked as a standalone "Prompt
context" row after the sent-message bubble. Tag steer messages as
session/steer:user and their context as session/steer:context, then
have classifyTurnItems wrap each steer message as a prompt segment
whose context rides the existing checks-icon dialog. Steer contexts
pair to steer messages positionally; orphans keep a visible standalone
row so metadata is never silently dropped. Prompt segment keys now
include the user message id since a turn can hold several prompt
segments (initial prompt + mid-turn steers).

Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
@tellaho
feat(desktop): collapse interleaved tool bursts into one summary row
bc83049 
Make the mixed pass interleave-tolerant: same-kind summaries produced
by the first pass now participate as burst children alongside leftover
eligible tool rows, so alternating investigation patterns (search →
read-summary → search → read-summary) collapse into a single
"Ran N tool calls" supervision row instead of many alternating rows.
Burst threshold drops to 2 segments. Mixed summaries carry ordered
child segments; nested same-kind summaries render intact and stay
expandable to their own rows, while flat leaf items preserve original
order for flattening and diff-stat aggregation. Messages, permissions,
failed tools, and status/suppressed rows still break bursts.

Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
@tellaho
fix(desktop): add vertical breathing room to the activity transcript …
85b244b 
…panel

The activity transcript in the agent session thread panel sat flush
against the bottom of the header row. Give the transcript container a
small y-padding (py-2) on the stacked child panel instead of touching
the global channel top-padding variable path.

Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
@tellaho
fix(desktop): flatten mixed tool summary children
9601b73 
Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
@tellaho
feat(desktop): animate grouped tool-call count increments
2e2a99b 
Route the leading count in tool-run summary labels ("Ran 16 tool
calls", "Read 3 files") through the existing AnimatedCount odometer so
streaming bursts roll their digits as the count grows. Gated on the
transcript animation preference; prefers-reduced-motion falls back to
static text via AnimatedCount's existing CSS, and the sr-only value
keeps the label readable for screen readers.

Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
@tellaho tellaho changed the title quiet agent activity tool bursts feat(agents): group activity tool bursts Jul 7, 2026
@tellaho tellaho changed the title feat(agents): group activity tool bursts feat(agents): group activity sequential tool Jul 7, 2026
@tellaho tellaho changed the title feat(agents): group activity sequential tool feat(agents): group activity sequential tools Jul 7, 2026
@tellaho tellaho changed the title feat(agents): group activity sequential tools feat(agents): group activity tool bursts Jul 7, 2026
@tellaho tellaho marked this pull request as ready for review July 7, 2026 08:53
@tellaho tellaho enabled auto-merge (squash) July 7, 2026 08:53
@tellaho tellaho changed the title feat(agents): group activity tool bursts feat(agents): group sequential activity tools Jul 7, 2026
wesbillman
wesbillman approved these changes Jul 7, 2026
View reviewed changes

@wesbillman wesbillman left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed by Pinky (Buzz agent) on Wes's behalf.

Summary: Two-pass activity transcript grouping — same-kind runs collapse to specific summaries ("Read 3 files"), then adjacent eligible routine tool rows of differing kinds (including pass-1 summaries) collapse into one mixed "Ran N tool calls" summary that expands directly to leaf tool rows, eliminating redundant nested summaries. Also routes steer prompt-context behind the steer bubble's ingress dialog, de-dupes prompt segment keys per user message id, and animates summary counts via the existing AnimatedCount.

Verified:

  • Guardrails: messages, permissions, failed tools (isError/error renderClass), status, and suppressed rows never join either grouping pass and break bursts — intervention points stay visible. Explicit render-class allowlist rather than a denylist. All test-pinned.
  • No dropped data: mixed summaries keep flat items in original order, flattenDisplayBlocks order preserved (tested), orphan steer context keeps its standalone row rather than being silently dropped.
  • Bounded recursion: mixed summary children flatten to leaf rows, and mixed summaries never re-enter a burst (isBurstParticipant).
  • splitActivityRowCountedObject is strict (rejects bare counts / trailing whitespace); reduced-motion and sr-only handled by the existing component — no new dependency.
  • Staleness: 8 commits behind main; only overlap is one className line in AgentSessionThreadPanel.tsx vs main's prop refactor — git merge-tree clean.
  • All CI checks green; full desktop unit suite 1902/1902 per PR validation.

Non-blocking notes:

  • The "File changes" detail block in the description lists ~38 files but the diff is 8 — looks like leftover text from a broader branch. Worth trimming so history isn't misleading.
  • The child.kind === "summary" render branch in SameKindSummaryItem is currently unreachable (mixed children are always flattened) — fine for type-completeness, but a comment or removal would tighten it.

@tellaho tellaho merged commit 426f044 into main Jul 7, 2026
45 of 47 checks passed
@tellaho tellaho deleted the tho/activity-row-grouping branch July 7, 2026 15:34
@wesbillman wesbillman mentioned this pull request Jul 7, 2026
Merged
tellaho added a commit that referenced this pull request Jul 7, 2026
@tellaho
Merge remote-tracking branch 'origin/main' into tho/composer-live-act…
1ed2d43 
…ivity

* origin/main:
  fix(desktop): hydrate reactions for Inbox context messages (#1596)
  fix: cleanup old screenshots that my agents committed (#1598)
  chore(release): release Buzz Desktop version 0.3.46 (#1585)
  fix(desktop): preserve agent model/provider when persona snapshot fields are blank (#1583)
  feat(acp,desktop): identify and reap stale agent harness processes (#1582)
  feat(desktop): active-draft badge, send-from-drafts confirm dialog, thread-deleted state (#1581)
  fix(desktop): treat baked build env vars as satisfying required agent config (#1580)
  feat(desktop): add "Copy image" to image right-click context menu (#1579)
  fix(nest): use buzz-dev symlink name for dev builds (#1587)
  fix(composer): address image-editor follow-up nits on #1491 (#1565)
  fix(desktop): render black static boot screen (#1570)
  feat(agents): group activity tool bursts (#1571)
  feat(desktop): aggregated overview rail, commit detail page, and full breadcrumbs (#1573)
  fix(desktop): fetch profiles for reaction actors and thread-reply authors (#1550)
  refactor(desktop): unify EditAgentDialog styling with PersonaDialog (#1540)
  feat(desktop): add 10-minute message grouping window (#1578)
  feat(desktop): unify sidebar section actions into a per-section ⋮ menu (#1577)
  feat(desktop): emoji avatar picker for agents + reliable picker scroll (#1576)

Co-authored-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
tlongwell-block pushed a commit that referenced this pull request Jul 7, 2026
Merge remote-tracking branch 'origin/main' into eva/community-moderation
faeb798 
* origin/main:
  docs(readme): add Getting started section routing install paths (#1606)
  fix(desktop): restrict shared-agent sync to dev data dirs (#1597)
  feat(desktop): restart-required badge from spawn-time config hash (#1602)
  feat(desktop): boot-time reconcile of managed agents to relay events (#1601)
  feat(desktop): canonical <PubKey> component — hover to view/copy full keys, owner "you" labels (#1589)
  fix(desktop): hydrate reactions for Inbox context messages (#1596)
  fix: cleanup old screenshots that my agents committed (#1598)
  chore(release): release Buzz Desktop version 0.3.46 (#1585)
  fix(desktop): preserve agent model/provider when persona snapshot fields are blank (#1583)
  feat(acp,desktop): identify and reap stale agent harness processes (#1582)
  feat(desktop): active-draft badge, send-from-drafts confirm dialog, thread-deleted state (#1581)
  fix(desktop): treat baked build env vars as satisfying required agent config (#1580)
  feat(desktop): add "Copy image" to image right-click context menu (#1579)
  fix(nest): use buzz-dev symlink name for dev builds (#1587)
  fix(composer): address image-editor follow-up nits on #1491 (#1565)
  fix(desktop): render black static boot screen (#1570)
  feat(agents): group activity tool bursts (#1571)
  feat(desktop): aggregated overview rail, commit detail page, and full breadcrumbs (#1573)
  fix(desktop): fetch profiles for reaction actors and thread-reply authors (#1550)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@wesbillman wesbillman wesbillman approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@tellaho @wesbillman