Skip to content

feat: reply threads on crewmate chat messages (ui) - #12815

Merged
iamwhatever merged 2 commits into
mainfrom
feat/reply-threads-ui
Sep 24, 2026
Merged

iamwhatever merged 2 commits into
mainfrom
feat/reply-threads-ui

Conversation

@CrysisDeu

@CrysisDeu CrysisDeu commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

A crewmate's chat is one flat list. When the user wants to ask about ONE thing the crewmate said ("what about the other 8?"), the question and its answers land in the main chat, between everything else. There is no way to discuss one message on its own, the way a Slack thread does.

Implements the accepted RFC: docs/request-for-change/rfc-crewmates-launch.md (screen 07, reply threads).

The CrewMates launch review decided screen 07: any bubble, yours or the crewmate's, can carry a reply thread that opens in the right side panel while the main chat stays visible. This PR is the frontend half; the API half, #12797, and the dashboard.crewmate_threads flag, #13162, are merged. It carries the one open PR it stacks on, #12923 (the grouped bubbles this panel reads its corner rule from), as ONE chore: carry … commit under its own, so every check runs against main; that commit goes away when #12923 lands. Review only the second commit.

Why it matters

Without threads, every follow-up question pushes the crewmate's real findings up and out of view. The main chat shows only what the crewmate says to you; side questions do not belong in it.

What changed (motivation → approach → change)

Piece What it does
ThreadFooter under a bubble faces of who took part, "N replies" in accent, "Last reply 2h ago" muted; one click opens the thread
"Reply in thread" in the action row on every user bubble and every finished crewmate reply (AssistantMessage, UserMessage); a mid-turn crewmate row has no Reply until the turn's last reply
pages/members/ThreadPanel the open thread in the side panel: quoted parent, hairline count, small bubbles, typing row, Reply composer
ThreadHooks on the render context only a crewmate's chat offers threads; every other surface draws nothing new
dashboard.crewmate_threads the flag (#13162, merged), default off; Settings → Developer → Crewmates toggle

Flagged, default off. hooks/useCrewmateThreadsFlag.ts reads the shared ['kirocrewConfig'] query: on only once a read said true; a failed read is its own failed state, an ErrorNotice with Retry beside the chat while the last known value stands, never shown as "off". Off, the Members page makes no footer read, passes no threads hooks, mounts no panel. The switch (pages/settings/CrewmatesSection.tsx, mounted by DeveloperPanel.tsx under the Crew Members preview card) PATCHes the key and reads it back through the same query; a failed read or save is an ErrorNotice with the agent hand-off (a toggle holds no draft), the read with a Retry; scripts/settingsExtract.ts maps the file so gen:settings indexes the row (settingsRegistry.gen.ts, src/kiro_crew/docs/settings-registry.generated.json). Out of scope: config has no crewmate parent flag (Crew Members is a client-side preview flag); the Crewmates heading is visual only.

Only the Members page hands components/ChatPane.tsx a threads hook set; ChatMessageList puts it on the MessageRenderContext, and renderAssistantBubble (app-sdk/messageRenderers.tsx, shared with #12923's run row in pages/chat/transcriptRenderers.tsx) and the user rows read it: a bubble whose mid has replies gets the footer, its action row the MessageSquare button. Without the hooks the rows are unchanged. The action row keeps two buttons: Copy and Raw markdown move into More, as when Speak is on.

Either opens the thread over the side panel's tabs (slides in; fades under prefers-reduced-motion) and hands them back on close (Escape too). The draft is kept per thread (threadDrafts). Inside: the quoted parent, a hairline count, the replies as small bubbles on the chat's run and corner rule (components/chat/crewmateBubbles.ts), a typing row, and a "Reply in thread…" composer (SendBtn disabled while a reply is in flight or the draft is over 32 KiB; the over-long hint is muted text).

Stored replies and footer data are React Query reads (api/threads.ts); the reply in progress streams through state/threadLiveStore.ts from chat.thread_reply frames (hooks/useWebSocket.ts); a stored frame refetches both, a reconnect every observed thread. A send's success clears the composer only if it still holds the sent text. Failures render through ErrorNotice, one plain sentence picked by the backend code: a failed send keeps the draft and its reply id (threadPendingReplies), so re-sending is the same reply to the server; a failed thread read has a Retry and the composer waits for it, muted; a failed footer read says the replies are still saved, with a Retry. Twenty-two strings under pages.chat.thread plus six under pages.settings.crewmatesSection, twelve catalogs and en-XA; one plural key in pluralKeys.json.

Carried commit: #12923's bubble run (crewmateBubbles.ts, CrewmateMessage.tsx, handlers/members.py, eventlog/members_projections.py, eventlog_hooks.py, preview_text.py, crew_log/, system_notices.py, history.py, history_projection.py, .github/black-baseline.txt, crew-mode.md, test/CrewmateMessage.test.tsx, the crewmate-chat-clean scenario and registry rows), cherry-picked onto main with four additive conflicts kept both ways (#12923 meets the same four on its rebase). It also restates one recency comment in history_projection.py (an Opus finding on #12923's code; #12923 takes it). Spec: docs/system-specs/modules/history.md, a sentence in crew-mode.md, a Reply threads row in docs/feature-map/README.md whose entry path starts at the switch; the nightly scenario (crewmate-reply-thread.yaml, features.json) turns it on first.

Backwards compatibility

Compatible: new optional props and context fields only; a surface that passes no threads hooks renders exactly as before. Everything is off until dashboard.crewmate_threads is turned on.

Tests

threadLiveStore.test.ts: deltas of one run accumulate, the terminal frame clears, another run replaces, an error is kept until the next reply, user frames are ignored, threads stay apart, reset drops every row, drafts and pending reply ids are kept per thread. ThreadFooter.test.tsx: the count in words, the last-reply time, the faces in first-appearance order, one button that opens the thread on its bubble's side. api/threads.test.ts: the three calls' URLs (slot and mid encoded), the reply body with and without the client reply id, a refusal as an ApiError with the backend code, stable query keys. ThreadPanel.test.tsx: the quoted parent, the count, the grouped crewmate run and the "You" label; Enter sends, the draft clears, send is disabled while the crewmate replies; the streamed reply shows and gives way to the stored one; a retry of the same text re-sends the same reply id, also after close and reopen, and an edited text gets a new one; a thread that failed to load takes no reply until it is retried and puts the keyboard on Retry; a 409 reads as a plain sentence and keeps the draft (a full chat points at a new chat, not a retry); the draft is measured in UTF-8 bytes, not characters; close hands the panel back; Escape closes and focus returns to the opener on unmount; the draft survives close and reopen and a sent reply clears it; a failed load shows the notice with a Retry that re-runs the read. threadLiveStore.test.ts also covers reset (every row dropped, every thread notified) and the per-thread drafts. en.context.json gains the translator note for the three-letter you label. CrewmatesSection.test.tsx: the flag hook is off until the config says true, the switch saves the key and refetches the shared config, a refused save and a failed read are said (the read with a Retry that re-enables the switch); ThreadPanel.test.tsx also: words typed while a send is in flight survive its success; MembersPage.test.tsx: off (the default) no footer read is issued and no thread notice or panel is drawn, on the read is issued for the confirmed slot and its failure is shown, a failed config read is a notice with a Retry that turns the feature on in place. CrewmatesSection.test.tsx also: the hook reports a failed read as failed (not off) and keeps the last known value through a failed refetch.

Agentic (gui-user) scenario, tier nightly: test/gui_user/scenarios/crewmate-reply-thread.yaml — a first-time user turns on the Crew Members preview, chats with Nova Sky, hovers the crewmate's reply, clicks Reply in thread, sees the Thread panel open beside the chat with the bubble quoted and "No replies yet", and sends a reply there; expectations: the panel is open with the chat still visible, the reply is in the panel, the bubble's footer reads "1 reply" ("2 replies" once the crewmate answered), and the main chat does not carry the reply. Registered in test/gui_user/features.json (+ regenerated FEATURES.md) and the shipped-scenario registry in test_scenarios_and_report.py. Not run locally; the nightly lane drives it.

Manual verification

Screenshots: real components rendered against stubbed /api (Vite + Playwright); backend exercised by CI only. The harness mounts the real MembersPage (roster, Radar's chat, side panel, the shipped ThreadFooter, "Reply in thread" action and ThreadPanel) with every /api answer, the thread endpoints included, intercepted by the capture script; each frame's state was asserted (footer text, revealed action row, panel heading, quoted parent, reply order, corner rule, composer, main chat visible beside the panel, "No replies yet" with the composer focused, Copy text and Raw markdown in the open menu, the typing row with send disabled, the load notice with its Retry, the send-failure notice with the draft still in the box, the Members-page summary-failure row with Retry over the mounted chat, and the footer badge refetched to the panel's count after a send) before the shot; the recording's first frame is checked non-blank. Harness and script are not committed.

Screenshots / video

The switch: Settings → Developer → Crewmates, off (the default) and on:

Settings Crewmates section, reply threads off, dark

Settings Crewmates section, reply threads on, dark

Settings Crewmates section, reply threads off, light

Footer under Radar's triage bubble ("4 replies", faces, last reply) and the hovered question bubble's action row with Reply in thread:

Thread footer and Reply in thread action, dark

Thread footer and Reply in thread action, light

The reply thread open in the side panel beside the chat: quoted parent, hairline count, replies on the corner rule, composer:

Thread panel open, dark

Thread panel open, light

Starting a new thread from Reply in thread on a message with no replies yet (parent quoted, "No replies yet", composer focused):

New thread, empty state, dark

New thread, empty state, light

The crewmate reply's More menu on this surface, holding the relocated Copy text and Raw markdown:

More menu with Copy text and Raw markdown, dark

More menu with Copy text and Raw markdown, light

A reply just sent: the stored row lands, "Radar is replying…" shows while the crewmate answers (send disabled), and the footer badge under the bubble has refetched to the same count the panel shows:

Typing row after a sent reply, dark

Typing row after a sent reply, light

The reply streaming in: two chat.thread_reply deltas so far, the growing bubble in place of the typing row, no stored crewmate row yet:

Reply streaming in, dark

Reply streaming in, light

A thread on the user's own bubble: the footer sits under the right-aligned bubble, Reply in thread on its hovered action row; opened, the panel quotes the user's message as the parent with Radar's reply first:

User bubble with footer and Reply action, dark

User bubble with footer and Reply action, light

Thread on the user's message open, dark

Thread on the user's message open, light

A send that failed: the notice above the composer names the affordance ("Press send to try again"), the draft kept, footer badge and panel count agreeing:

Send failed, draft kept, dark

Send failed, draft kept, light

The per-slot footer read failed: the Members page says so above the chat ("Your replies are still saved; the chat itself is up to date."), with Retry, and the chat stays mounted:

Reply-thread summary failed, dark

Reply-thread summary failed, light

A thread that failed to load: the notice with Retry (focus lands there), the composer muted and disabled with "Press Retry to load the thread, then reply." in place of its placeholder, the draft kept:

Thread load failed with Retry, composer waiting, dark

Thread load failed with Retry, composer waiting, light

Open from the footer, close with Escape (recording starts on the painted chat):

Thread open and close

shots 01–02 recorded from 16a5c4a (before the panel moved onto #12923's crewmateBubbles module: same surface and corner rule, bubble padding now the chat's px-3.5, 2px wider than shown); the panel stills (composer now reads "Reply in thread…") and the recording from the round-24 tree, the More-menu still from round 14 · feat/reply-threads-ui worktree · mode: Vite dev + Playwright, /api stubbed · real components, stubbed API (no backend), no fixtures in the commit

Related Issues

no linked issue: CrewMates launch screen 07 (reply threads), frontend half; depends on #12923 (#12797 and #13162 are in main).

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

@CrysisDeu

Copy link
Copy Markdown
Contributor Author

Intent: Let the user open a reply thread on any message of a crewmate's chat — a footer under bubbles that have replies, "Reply in thread" on every bubble, and the thread itself in the right side panel with a Reply composer — while the main chat stays visible.
Not a goal: Threads on ordinary (non-crewmate) chats, tool actions from inside a thread, the main chat's grouped-bubble rendering (a separate PR; this one carries a local corner-rule helper until then), or the backend contract (#12797, carried here until it lands).

@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 22, 2026
@github-actions

github-actions Bot commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — ✅ PASS

Design-level review of bbb2078f3fc0f91db7ad79a9ba5fbc069c471340 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Verify the carried #12923 commit is dropped by rebasing onto merged #12923 before this merges — otherwise ~2,500 unreviewed backend lines land under a chore: title.

[DESIGN-REVIEWED] bbb2078

@github-actions

github-actions Bot commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Fable 5) — 🟡 CONCERNS

UX-level review of bbb2078f3fc0f91db7ad79a9ba5fbc069c471340 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: CONCERNS

Thread entry points are recognized only by inference: the blind reader rated both the hover speech-bubble icon and the reply-count footer "a guess."

Watch

  • Both primary entry points were read correctly but only as guesses: the icon-only MessageSquare action ("the bubble likely starts a reply thread… a guess") and the footer ("a marker that this message has a side-conversation… a guess"). Every first-time user meets them; friction not failure, fading once learned. Smallest fix: add a labeled Reply in thread item to the More menu — the reader looked there and noted "this menu has no 'Reply' or thread option in it" (shot-10).
  • Two composers on screen at once: the reader distinguished "Message Radar…" from "Reply in thread…" but said "at a glance it would be easy to type in the wrong one" — a thread reply misdelivered to the main chat is recoverable but defeats the feature's purpose. Consider visually receding the main composer while the thread composer holds focus.

[UX-REVIEWED] bbb2078

@CrysisDeu
CrysisDeu force-pushed the feat/reply-threads-ui branch from e920798 to 6ceabed Compare September 22, 2026 13:26
@github-actions

github-actions Bot commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of bbb2078f3fc0f91db7ad79a9ba5fbc069c471340 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] bbb2078

False positive or not applicable? A repository writer can comment:
/ai-review override gpt bbb2078f3fc0f91db7ad79a9ba5fbc069c471340: <one-sentence reason>

@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ⚠️ review incomplete

GPT 5.6 did not produce a complete verdict for 6ceabed4fd20f947de3cb617f416e8eddae24145; inspect the workflow logs and re-run it.

This comment is updated in place on each push.

See the GPT 5.6 Review job logs; this commit has no completed GPT verdict.

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 6ceabed4fd20f947de3cb617f416e8eddae24145: <one-sentence reason>

@CrysisDeu
CrysisDeu force-pushed the feat/reply-threads-ui branch from 6ceabed to 724debd Compare September 22, 2026 14:37
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Sep 22, 2026
@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • Thread turns perform filesystem and SQLite reads on the event loop (span=d2196525dad2)

fixed in f2319e0 (the api commit this PR carries; this head is 724debd): KiroCrewConfig.load runs through asyncio.to_thread. resolve_agent_bindings stays inline on purpose, as the main chat's and the side chat's calls do -- offloading the resolver swallows a StopIteration into a Future and hangs the await; it reads only the warmed in-memory snapshot (warm_project_agent_names runs first, off-loop).

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • Load failure lacks the required hand-off decision (span=49f375a69f2e)

fixed in 724debd: the load-failure ErrorNotice in ThreadPanel carries its "No hand-off" decision comment naming the unsaved reply draft in the composer below; the read retries on the next chat.thread_reply frame or reopen.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • Reply threads are absent from the feature map (span=e7ac5689c151)

fixed in 724debd: docs/feature-map/README.md gains a "Reply threads" row naming the thread UI files (ThreadPanel, ThreadFooter, the row action, ThreadHooks, threadLiveStore, api/threads.ts), the handler (chat_threads.py, history.py, ws.py) and the three routes plus the chat.thread_reply frame; the Feature Map gate passes on this head.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • Reply creates a third peer action in the message row (span=38644d4f5240)

fixed in 724debd: on the thread surface (onReplyInThread set) the assistant row keeps two peer controls -- Reply and More -- and Copy plus the raw-view toggle move into the More menu, the same swap the row already makes for Speak (copyInMenu); every other surface is unchanged. The user row on that surface holds Copy and Reply, two.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 22, 2026
@github-actions

github-actions Bot commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of bbb2078f3fc0f91db7ad79a9ba5fbc069c471340 — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: PASS

Verify the undeclared Work-log "chats"→"runs" relabel in every locale is intended to ship here rather than with #12923.

Not justified as shipped

  • Item 7 — undeclared: the readable description never mentions relabelling activity_chat_count ("1 chat" → "1 run") across 13 catalogs; the reason lives only in the carried crew-mode.md hunk ("counts the member's runs … so it does not contradict a chat that just said the crewmate has not spoken").
  • Item 8 — rides along: the whole docs: record the crewmate chat contract #12923 base (speech-only crewmate chat, roster preview reconcile) is in this diff as the declared chore: carry commit; docs: record the crewmate chat contract #12923 owns its review.

What this change ships

Inventory (10 items, capped) — 8 justified

Intent: let a user discuss one message of a crewmate's chat in a Slack-style side-panel thread (frontend of RFC screen 07) — an ADDITION, flagged off.

  1. "Reply in thread" on user bubbles and finished crewmate replies (flag on) — justified
  2. Replies footer under a threaded bubble: faces, count, last-reply time — justified
  3. Thread panel over the side-panel tabs: quoted parent, small bubbles, typing row, composer, per-thread drafts, Escape/focus return — justified
  4. Settings → Developer → Crewmates toggle for the merged dashboard.crewmate_threads flag, default off — justified
  5. Copy and Raw markdown move into More on bubbles carrying Reply — justified
  6. Live reply streams over chat.thread_reply; a reconnect refetches every observed thread — justified
  7. Work-log activity counts relabelled "chats" → "runs" with a defining tooltip, all locales — undeclared (derived reason recorded only in the carried spec)
  8. Carried docs: record the crewmate chat contract #12923 base: speech-only crewmate chat runs, roster preview reconcile — rides along (declared stack carry, reviewed on docs: record the crewmate chat contract #12923)
  9. Twenty-eight strings in 12 catalogs + en-XA, one plural key — justified
  10. Two nightly GUI scenarios and a feature-map row — justified

[FIRST-PRINCIPLES-REVIEWED] bbb2078

@github-actions

github-actions Bot commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5 Review — ✅ no blocking findings

Reviewed bbb2078f3fc0f91db7ad79a9ba5fbc069c471340 — this comment is updated in place on each push.

Review details

Two advisory findings; nothing blocks the merge.

FINDING — src/kiro_crew/dashboard/handlers/members.py:463 — the three new continues sit above if stopped: row["last_message_stopped"] = True, so a slot with in-memory rows past the last flush (the ordinary state for the ~5s after the user presses Stop, and after _resolve_stop_event/_rearm_stop_event set slot._dirty) or a non-exhaustive empty speech read returns a row with the flag absent, and MembersPage.tsx:533 {view.last_message_stopped && (…)} draws no chip — the merged view has fallbacks for last_message/last_active_ts but none for this field, so a just-stopped thread reads on the roster as ongoing work → Fix: move if stopped: row["last_message_stopped"] = True up next to row["last_message"] = preview, leaving only preview_authoritative.add(row["slot_key"]) behind the guards.

FINDING — website/src/pages/members/ThreadPanel.tsx:208 — const unreadable = detail.isError is true on a failed refetch that still holds detail.data (a chat.thread_reply terminal frame invalidates threadQueryKey, retryPolicy retries once, then status flips to error with the cached thread intact), which sets disabled on thread-composer, makes submit() return early, and fires the [unreadable] effect that pulls focus to thread-load-retry out from under a half-typed reply, while the parent and every stored reply stay on screen → Fix: const unreadable = detail.isError && !detail.data.

[OPUS-REVIEWED] bbb2078

Verdict parsed from the review's SHA-scoped output markers for commit bbb2078f3fc0f91db7ad79a9ba5fbc069c471340.

False positive or not applicable? A repository writer can comment:
/ai-review override fable bbb2078f3fc0f91db7ad79a9ba5fbc069c471340: <one-sentence reason>

@CrysisDeu
CrysisDeu force-pushed the feat/reply-threads-ui branch from 724debd to 1bf49af Compare September 22, 2026 15:22
@CrysisDeu

CrysisDeu commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor Author
  • threadsQuery.isError is never surfaced (span=e7ac5689c151)

fixed in 1bf49af: MembersPage renders an inline ErrorNotice above the chat when the per-slot footer read fails ("Couldn't load the reply threads for this chat. The chat itself is up to date.") with a Retry beside it that refetches in place; no hand-off, because the DM composer below holds an unsaved draft.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • _in_flight.add lacks an encompassing finally; an OSError from the store write leaks the reservation (span=d9f47990731d)

fixed in aecbba9 (the api commit this PR carries): the reservation is released by one finally on every path that does not hand it to the turn -- each refusal and any exception the store write raises, an OSError included. Pinned by test_a_store_error_nothing_names_still_releases_the_mark (append_thread_reply raises OSError(28) -> 500, mark released).

@CrysisDeu

CrysisDeu commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor Author
  • Screenshots missing (span=117a852ee21b)

pending: the pod fleet this worktree's screenshots come from is down fleet-wide (pod_up answers repo_unavailable; the operator is on it). The PR stays a draft until the dark and light captures of the footer + hover action and of the open thread panel are attached; they are attached before it is marked ready, never committed.

@CrysisDeu

CrysisDeu commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor Author
  • err_too_long names no next step (span=f20db36a0f8c)

fixed in 1bf49af: "That reply is too long. Shorten it to 32 KB or less." in all twelve catalogs.

@CrysisDeu

CrysisDeu commented Sep 22, 2026 •

Copy link
Copy Markdown
Contributor Author
  • Side Chat premise (span=4d361c396f6f)

rebutted: a reply thread is not the side chat. The side chat is an off-record conversation about the whole chat that is never stored with it; a thread is on-record replies attached to ONE message, stored beside the transcript and shown under that message to whoever opens the chat later. What they share -- the isolated session and the read-only tool posture -- is shared code, and the screen is the CrewMates launch's decided screen 07 (the api half asked the maintainer to confirm it on #12797).

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • Reply action absent when showFooter is false, contrary to the feature-map claim (span=2f6082500a24)

fixed in 1bf49af: the claim is narrowed to what the code does -- the feature-map row and the PR body now say every user bubble and every FINISHED crewmate reply carry Reply in thread, and that a mid-turn crewmate row running on into a tool call has no footer row and so no Reply until the turn's last reply. The footer row is where a finished reply's actions live; a second action surface for mid-turn rows would be a new control the design does not have.

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • Draft is lost on close (span=49f375a69f2e)

fixed in 1bf49af: the composer's text is kept per slot+mid in threadDrafts (state/threadLiveStore.ts, memory only -- the main composer's draft is not persisted across reloads either), read back on mount and cleared by a sent reply. Pinned by "keeps the draft across close and reopen, and a sent reply clears it".

@CrysisDeu

Copy link
Copy Markdown
Contributor Author
  • A final frame lost during a disconnect leaves a partial reply forever (span=88cac34add2d)

fixed in 1bf49af: on WS reconnect useWebSocket calls threadLiveStore.reset() (every live row dropped, every subscriber notified) and invalidates the ['chat-thread'] and ['chat-threads'] prefixes, so the stored replies -- the truth -- are refetched for every observed thread and footer. Pinned by "reset drops every held row and notifies each thread".

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Sep 23, 2026
@CrysisDeu
CrysisDeu force-pushed the feat/reply-threads-ui branch from a3df9e0 to 56233d5 Compare September 23, 2026 02:16
@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ⚠️ review incomplete

GPT 5.6 did not produce a complete verdict for a3df9e063fa7c7a145cd064d63fa650268c4606b; inspect the workflow logs and re-run it.

This comment is updated in place on each push.

See the GPT 5.6 Review job logs; this commit has no completed GPT verdict.

False positive or not applicable? A repository writer can comment:
/ai-review override gpt a3df9e063fa7c7a145cd064d63fa650268c4606b: <one-sentence reason>

@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ⚠️ review incomplete

GPT 5.6 did not produce a complete verdict for 56233d5f37132ad90acd3824738ab0c17349c06d; inspect the workflow logs and re-run it.

This comment is updated in place on each push.

See the GPT 5.6 Review job logs; this commit has no completed GPT verdict.

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 56233d5f37132ad90acd3824738ab0c17349c06d: <one-sentence reason>

@CrysisDeu
CrysisDeu force-pushed the feat/reply-threads-ui branch from 56233d5 to 2009068 Compare September 23, 2026 02:41
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Sep 23, 2026
@CrysisDeu
CrysisDeu force-pushed the feat/reply-threads-ui branch from 2009068 to c1de825 Compare September 23, 2026 03:23
@github-actions github-actions Bot removed the readiness: action required A blocking check or review needs attention label Sep 23, 2026
@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — 🔴 changes requested (blocking)

GPT 5.6 found at least one blocking issue that must be resolved before merging c1de8251cacbc0e853461bb6ab20bcf4cd1ce048. Adjudication produced no [GPT-ADJUDICATED] marker for c1de825: the adjudication stage FAILED TO RUN (its raw output is in the Adjudication details block -- typically an API or setup error), so no proportionality review took place and GPT's blocking verdict stands unreviewed.

This comment is updated in place on each push.

BLOCKING -- docs/feature-map/README.md:107 -- Legacy messages cannot start claimed threads

“a reply thread on any one message”
Existing row without meta.mid -> threadMidOf returns undefined -> Reply action is absent.
Anchor: feature-map-correctness
Fix: Limit the claim to messages with durable IDs.

BLOCKING -- website/src/pages/members/ThreadPanel.tsx:131 -- Lost responses can duplicate durable replies

mutationFn: (text: string) => threadsApi.reply(slot, mid, text),
Accepted POST loses its response -> mutation retains draft and prompts retry -> retry stores a duplicate reply and starts another turn.
Anchor: residual/crash-data-loss-corruption
Fix: Add a client request ID and deduplicate repeated IDs server-side.

[BLOCK-MERGE] c1de825
[GPT-REVIEWED] c1de825
False positive or not applicable? A repository writer can comment:
/ai-review override gpt c1de8251cacbc0e853461bb6ab20bcf4cd1ce048: <one-sentence reason>

@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — 🔴 changes requested (blocking)

GPT 5.6 found at least one blocking issue that must be resolved before merging e8f5f0b93629a5495cf0907834906f49b3943fde. Adjudication produced no [GPT-ADJUDICATED] marker for e8f5f0b: the adjudication stage FAILED TO RUN (its raw output is in the Adjudication details block -- typically an API or setup error), so no proportionality review took place and GPT's blocking verdict stands unreviewed.

This comment is updated in place on each push.

BLOCKING -- src/kiro_crew/history_projection.py:1304 -- Thread deletion follows a symlinked sidecar directory (origin: validation)

if threads_path.exists(): ... os.replace(threads_path, threads_staged) / os.lstat(sidecar)
Symlinked .threads directory -> session deletion or reclaim -> external matching file is moved and purged.
Anchor: residual/security
Fix: Refuse linked parents and use descriptor-relative, no-follow operations in both staging paths.

BLOCKING -- src/kiro_crew/dashboard/chat_threads.py:540 -- A lost 202 followed by a failed turn makes retry a no-op

if existing.get("id") == reply_id: return ... duplicate ...
Lost response -> failed background turn stores no answer -> retained-ID retry returns success without launching and clears the draft.
Anchor: residual/crash-data-loss-corruption
Fix: Relaunch an unanswered matching reply when no turn is active; return duplicate success only when active or answered.

[GPT-REVIEWED] e8f5f0b
[BLOCK-MERGE] e8f5f0b
False positive or not applicable? A repository writer can comment:
/ai-review override gpt e8f5f0b93629a5495cf0907834906f49b3943fde: <one-sentence reason>

@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ⚠️ review incomplete

GPT 5.6 did not produce a complete verdict for 5efdeda52a344c616b0c62f4eda0a03511ea24f3; inspect the workflow logs and re-run it.

This comment is updated in place on each push.

See the GPT 5.6 Review job logs; this commit has no completed GPT verdict.

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 5efdeda52a344c616b0c62f4eda0a03511ea24f3: <one-sentence reason>

@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — 🔴 changes requested (blocking)

GPT 5.6 found at least one blocking issue that must be resolved before merging aa5ca1725fdca3c5074b9c4e0fe553f235286085. Adjudication produced no [GPT-ADJUDICATED] marker for aa5ca17: the adjudication stage FAILED TO RUN (its raw output is in the Adjudication details block -- typically an API or setup error), so no proportionality review took place and GPT's blocking verdict stands unreviewed.

This comment is updated in place on each push.

BLOCKING -- src/kiro_crew/history.py:2145 -- Empty threads bypass the sidecar item budget
budget -= len(kept) / out[mid] = kept
A 64 MiB sidecar of valid mids mapped to [] -> read_threads() retains every key without consuming its budget -> gateway memory exhaustion and crash.
Anchor: a-bound-bounds-every-field-it-retains
Fix: Skip empty lists or charge every retained key against the shared budget.

BLOCKING -- src/kiro_crew/session_storage.py:1384 -- Pinned sidecar publication is finalized through an unpinned path
os.link(src, dst.name, dst_dir_fd=dst_dir_fd)
Replace .threads after its fd is opened -> publication uses the old pinned directory but finalization fsyncs the replacement path and deletes the source -> restore succeeds with the sidecar absent from its canonical location.
Anchor: residual/crash-data-loss-corruption
Fix: Use dst_dir_fd for metadata, durability, cleanup, and rollback until the source is dropped.

[BLOCK-MERGE] aa5ca17
[GPT-REVIEWED] aa5ca17
False positive or not applicable? A repository writer can comment:
/ai-review override gpt aa5ca1725fdca3c5074b9c4e0fe553f235286085: <one-sentence reason>

@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ⚠️ review incomplete

GPT 5.6 did not produce a complete verdict for 6f37b688410c075f1ca1b596905019c30d75d233; inspect the workflow logs and re-run it.

This comment is updated in place on each push.

See the GPT 5.6 Review job logs; this commit has no completed GPT verdict.

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 6f37b688410c075f1ca1b596905019c30d75d233: <one-sentence reason>

iamwhatever
iamwhatever previously approved these changes Sep 24, 2026
W4's feat/crewmate-chat-bubbles head 955d1f5, cherry-picked onto main
e183a46 with four additive conflicts kept both ways (feature-map row,
crew-mode section, ChatPane renderer options, transcriptRenderers option
types). #13162 (threads flag) is in main and is dropped from the carry.
Not this PR's review surface; merges with #12923.
Under a bubble that has replies, a footer: the faces of who took part,
"N replies" in accent, "Last reply 2h ago" muted. Every bubble's hover
action row, the user's included, gets "Reply in thread". Either opens
the thread in the right side panel: the parent quoted as one bubble, a
hairline reply count, the replies as small bubbles on the grouped-corner
rule (crewmate left, user right-aligned), a typing row while the
crewmate replies, and a one-line Reply composer. The main chat stays
visible beside it; closing hands the panel's tabs back.

Only a crewmate's chat offers threads: the Members page hands ChatPane a
ThreadHooks set the assistant and user rows read off the render context;
every other surface draws neither footer nor action. Stored replies and
the per-slot footer data are React Query reads; the reply in progress
streams through a small live store fed by chat.thread_reply frames.
@bolichen97

Copy link
Copy Markdown
Contributor

Maintainer sweep (2026-09-24, on behalf of @bolichen97) — decision: holding until #12923 lands. This PR itself passes every gate (readiness: passed, PR Readiness green, mergeable, no blocking review), but its head is a 2-commit stack whose first commit 3b07cec chore: carry #12923 (bubbles) as this stack's base carries ~2,500 backend lines from the still-open #12923 (currently readiness: action required, Coverage Gate failing, prior approval dismissed). Squash-merging now would land that code under a chore: title without its own review. @CrysisDeu — once #12923 merges, rebase this onto main so the carry commit disappears; auto-merge is already armed by @iamwhatever, so a re-approval is all it will need. If you prefer, I can approve #12923 first once its readiness goes green.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants