feat(agents): show per-turn duration and prune dead turns within ~25s of host crash#1017
Merged
Conversation
The Working badges showed only which channels an agent was active in, discarding the per-turn start time the store already tracked. Each badge now reads "Working in #channel · 1m 23s" and ticks once a second. Elapsed is anchored to a desktop-clock observedAt captured when a turn enters the store, not the observer event's startedAt — the latter is the agent host's clock, which skews against the desktop's Date.now() for remote agents and would make the counter start negative or jump. The 1s tick lives in a leaf WorkingBadge so only visible badges re-render each second; idle rows never mount the ticking hook. Co-authored-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
The same-channel collapse must read the live turn map on every mutation, not a stale cached minimum. Ending the earliest-observed turn must advance the surfaced observedAt to the surviving turn — pinned with a deterministic test (busy-wait guarantees distinct observedAt across the clock-millisecond boundary). Adds a WHY comment at the observedAt capture site so a future reader does not revert it to startedAt and reintroduce cross-machine skew. Co-authored-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
The desktop prunes a turn after 90s of no observer activity, which only matters when an agent host dies without unwinding (kill -9 / crash) so the turn_completed Drop guard never fires. With no liveness signal the badge can linger 90s. The harness now emits a turn_liveness observer event per in-flight turn on a fixed interval (BUZZ_ACP_TURN_LIVENESS_SECS, default 10, 0=off) so the desktop can prune dead turns far sooner. Named turn_liveness, not heartbeat — heartbeat already denotes agent self-prompting in this crate. The emission rides every prompt-await path as a non-resolving select! arm, so it lives and dies with the prompt future and stops on every exit path with no separate teardown. It uses a captured ObserverHandle rather than agent.acp because the prompt holds &mut agent.acp for its whole duration. Co-authored-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
The desktop pruned a turn after 90s of silence, which only bites when an agent host dies without unwinding so the turn_completed Drop guard never fires. The harness now emits turn_liveness pings (Phase 1); routing them through the same recordActivity path as acp_read/acp_write lets the bound drop to 25s (2.5x the 10s liveness interval) without falsely pruning live-but-quiet turns. The bound is derived from LIVENESS_INTERVAL_MS so it tracks the interval if it changes. Co-authored-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Phase 2 dispatch named three new tests; the keep-alive and prune-at-bound cases shipped but the null-turnId no-op was missing. recordActivity early-returns on a falsy turnId, so a liveness ping without a turn_id must refresh nothing and must not throw — proven via the turn still pruning at the bound after the null ping. Co-authored-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Co-authored-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Collaborator
Author
Single agent working — live counterPaul mid-turn in #general: the
Multiple channels — independent countersPaul working in two channels at once —
Longer-running turn — h/m/s rolloverA turn well past the hour mark (
|
wpfleger96
pushed a commit
that referenced
this pull request
Jun 12, 2026
This was referenced Jun 12, 2026
tlongwell-block
pushed a commit
that referenced
this pull request
Jun 13, 2026
* origin/main: (33 commits) fix(desktop): make Windows release compile cleanly (#1029) Add production Docker Compose bundle (#985) feat(profile): show active turn badges on agent profile panel and popover (#1026) chore(release): release version 0.3.20 (#1027) fix(release): resolve Windows sidecar path and Linux AppImage updater format (#1024) chore(release): release version 0.3.19 (#1014) fix(release): ignore prerelease tags in changelog generation (#1021) fix: repair main build after cross-PR merge skew (#1020) feat(agents): show per-turn duration and prune dead turns within ~25s of host crash (#1017) fix(release): replace hermit with native tool setup on Windows job (#1018) feat(acp): surface error-class outcomes to the activity feed only, never the channel (#1010) fix(desktop): migrate Sprout workspace storage (#1016) feat(auth): force token refresh on rejected token (401/403), never the browser (#1015) fix(release): mark prerelease versions so they do not become latest (#1013) feat(acp): implement systemPrompt with protocol version gating (#981) fix(release): update repository name check from block/sprout to block/buzz (#1012) feat(release): all-OS desktop builds + universal auto-update manifest (#1011) Add relay disconnect UX: friendly errors, reconnect, cached identity (#1004) feat(agents): add active turn indicators to Agents Menu (#1005) ci: add fork guards to docker, release, and auto-tag workflows (#1007) ... Co-authored-by: npub1t2tgm7d8f995uqvmnm8h88sg3wnpp9a5xysjf6dg3tjmgt3ltulqdp8ehr <5a968df9a7494b4e019b9ecf739e088ba61097b4312124e9a88ae5b42e3f5f3e@sprout-oss.stage.blox.sqprod.co> Signed-off-by: npub1t2tgm7d8f995uqvmnm8h88sg3wnpp9a5xysjf6dg3tjmgt3ltulqdp8ehr <5a968df9a7494b4e019b9ecf739e088ba61097b4312124e9a88ae5b42e3f5f3e@sprout-oss.stage.blox.sqprod.co>
tellaho
pushed a commit
that referenced
this pull request
Jun 14, 2026
…tate * origin/main: (21 commits) fix(release): use signed NSIS installer for updates (#1036) handoff: pass full session history to summarizer (#1033) feat(emoji): latest-set-wins union for custom emoji across desktop, mobile, and CLI (#989) Fix relay NIP-11 software URL (#1030) fix(desktop): make Windows release compile cleanly (#1029) Add production Docker Compose bundle (#985) feat(profile): show active turn badges on agent profile panel and popover (#1026) chore(release): release version 0.3.20 (#1027) fix(release): resolve Windows sidecar path and Linux AppImage updater format (#1024) chore(release): release version 0.3.19 (#1014) fix(release): ignore prerelease tags in changelog generation (#1021) fix: repair main build after cross-PR merge skew (#1020) feat(agents): show per-turn duration and prune dead turns within ~25s of host crash (#1017) fix(release): replace hermit with native tool setup on Windows job (#1018) feat(acp): surface error-class outcomes to the activity feed only, never the channel (#1010) fix(desktop): migrate Sprout workspace storage (#1016) feat(auth): force token refresh on rejected token (401/403), never the browser (#1015) fix(release): mark prerelease versions so they do not become latest (#1013) feat(acp): implement systemPrompt with protocol version gating (#981) fix(release): update repository name check from block/sprout to block/buzz (#1012) ... Co-authored-by: Taylor Ho <taylorkmho@gmail.com> Signed-off-by: Taylor Ho <taylorkmho@gmail.com> # Conflicts: # desktop/src/features/profile/ui/UserProfilePanel.tsx
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Extends the Agents panel "working" indicator to show how long each agent has been running in its current turn, with a live-ticking counter per channel.
Producer — harness liveness emission (
crates/buzz-acp/)The desktop previously kept a turn's "working" badge alive for 90s after the last stream event, so a hard-killed agent host (
kill -9, crash — no graceful unwind) left a stale badge for up to 90s. The harness now emits aturn_livenessobserver event on a fixed interval while a turn is in-flight, carrying the liveturn_id. Because it is emitted harness-side and independent of agent output, it stops the instant the host process dies — which is exactly the signal the desktop needs to prune a dead turn quickly.run_turn_livenessis a non-resolving future raced against the prompt as aselect!arm on both the cancellable and non-cancellable paths, so it lives and dies with the prompt future and tears down on every exit path (complete / cancel / error / panic) — no separate cleanup to drift out of sync.ObserverHandle, not the borrowed acp client.BUZZ_ACP_TURN_LIVENESS_SECScontrols the interval (default10;0disables; floor of5smirrors the existing heartbeat-interval validation).Consumer — desktop backstop (
desktop/src/features/agents/activeAgentTurnsStore.ts)turn_livenessroutes through the same activity path asacp_read/acp_write, refreshinglastActivityAt.90_000to a derivedLIVENESS_INTERVAL_MS * 2.5(~25s), tolerating one fully dropped ping plus slack. Only a hard-killed host reaches this bound; graceful exits clear viaturn_completedand working turns refresh on stream events.Naming
The new concept is
turn_liveness, distinct from the existing agent self-promptheartbeat(PromptSource::Heartbeat,BUZZ_ACP_HEARTBEAT_INTERVAL) and the ACP-stream-layerkeepalive— both already taken for unrelated mechanisms.