FE-710: Chat runtime — unified surface with threads as primitive

[kostandinang]

Summary

Delivers the FE-710 chat-runtime-threads frontier — Track 2 of the Conversational Workspace Runtime umbrella.

Substrate (D153): new thread table sits between chat and turn (option q). chat becomes a pure container; turn.thread_id replaces turn.chat_id; chat.kind and chat.active_turn_id retire. Five thread kinds: interview / side / reconciliation / qa / agent_run. Flat threads via thread.invoked_in_turn_id (D154; no nested threads in V1).

Rendering: ThreadCollapsible interleaves non-interview threads inline after their invoking turn. Side threads carry live SSE streaming, optimistic local state, mode toggle (explore / edit → Ask / Edit per brief), and in-thread patch staging. Turn-zero kickoff turns are created when threads open.

Cutover: SideChatPopover deleted (−3,300 LOC across 5 files); side-chat-host shrunk 940 → 95 LOC. Eager thread creation route + openFor always routes inline.

Design brief: docs/design/UNIFIED_CHAT_UX.md — locked baseline for the unified chat UX (modes / mentions / layout states / kickoff copy / motion vocabulary). UX-layer items beyond what shipped here become new frontiers.

Acceptance

• thread table with five kinds
• turn.thread_id replaces turn.chat_id; chat.kind / chat.active_turn_id retire
• Threads render inline as collapsibles (interactive for side; scaffolded for others)
• SideChatPopover retires as cutover
• In-thread mutation state in ThreadCollapsible (criterion reframed — the global PatchListOverlay strip remains as a deliberate cross-thread summary surface, not a regression; in-thread mutation state is the per-thread replacement)
• Turn-zero universal entry across kinds
• Agent runs render inline via thread.invoked_in_turn_id

Known regression — Chat-from-item entry bridge

Clicking "Chat" on a knowledge item in structured-list view (and likely graph view) creates a side thread successfully but produces no visible feedback on the graph view itself. The thread is created and visible when the user navigates to the chat/transcript view.

Partially fixed: free-floating threads (null invoked_in_turn_id) now render at the end of the transcript instead of being filtered out. The remaining cause is that the graph/structured-list view doesn't render thread surfaces — <ThreadCollapsible> only mounts in the transcript view.

Resolves with: the UX layout system from UNIFIED_CHAT_UX.md §4 (Compact / Side-docked / Maximize / Full). The layout system provides a chat surface on every view, eliminating the need to navigate. Not blocking the substrate close.

Deferred to follow-up frontiers

• PendingReviewSection retirement → reconciliation-runtime (Track 3)
• Mention chip + autocomplete (# / $ / !)
• Layout state header control (Compact / Side-docked / Maximize / Full); default = Side-docked
• "Reconcile Now" sidebar affordance
• Reconciliation thread in-stream UX (target-grouped, classifier states)
• QA thread composer UX
• Ladle design prototype (scenes A / B / C / D)

References

• Linear: FE-710
• Design brief: docs/design/UNIFIED_CHAT_UX.md
• Substrate decisions: memory/SPEC.md — Req 45, D153 / D154, I111
• Umbrella: docs/design/CONVERSATIONAL_WORKSPACE_RUNTIME.md §3.2
• PLAN.md frontier: chat-runtime-threads (moved to Recently Completed)

Test plan

• npm run verify passes
• Manual: create a spec, advance through grounding turns, open a side thread on an item, send messages, verify streaming + persistence
• Manual: edit an item with impact > soft, verify staged patches appear in-thread
• Manual: refresh mid-stream, confirm history reconstructs correctly
• Manual: agent-run thread renders inline with Sparkles chip when invoked
• Known regression — "Chat" from structured-list view does not surface the thread (see above)

🤖 Generated with Claude Code

[cursor]

PR Summary

High Risk
High risk because it changes core persistence schema (thread table, turn.thread_id, chat reshaping) and removes the existing side-chat UI/bridging logic, which can impact data migrations and chat/edit flows across the app.

Overview
Delivers the thread substrate by adding a new thread table (with interview/side/reconciliation/qa/agent_run kinds), migrating turn records to reference thread_id, and simplifying chat to a container-only table; turn.phase is also made nullable for non-interview threads.

Cuts over the UI away from SideChatPopover by deleting the popover/host tests and removing the overlay scoping/undo bridge (PatchListOverlayBridgeProvider, PatchListUndoProvider), leaving PatchListOverlay to always apply/undo globally and shrinking SideChatHost to an inline-thread launcher that creates/focuses side threads.

Updates planning/spec docs (memory/SPEC.md, memory/PLAN.md) and adds a new UX design brief (docs/design/UNIFIED_CHAT_UX.md) to define the unified inline-thread chat surface.

^{Reviewed by Cursor Bugbot for commit 5a9ef97. Bugbot is set up for automated code reviews on this repo. Configure here.}

[kostandinang]

• FE-710: Chat runtime — unified surface with threads as primitive #138 👈 (View in Graphite)
• FE-709: Continuous workspace shell #134 : 1 other dependent PR (#139 )
• FE-705: Planning extensions and persistence facade cleanup #133 : 1 other dependent PR (#136 )
• FE-705: Agent CLI capabilities and workflow documentation #132
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[augmentcode]

🤖 Augment PR Summary

Summary: This PR updates the planning/spec artifacts to settle the chat-runtime-threads substrate decision (option (q): a new thread table between chat and turn) and track the work under FE-710.

Changes:

• Moves chat-runtime-threads to memory/PLAN.md Active and links it to FE-710, updating objective/acceptance/traceability.
• Adds an explicit “current execution pointer” describing the initial substrate-landing slice (schema + migration, no UI cutover).
• Extends memory/SPEC.md with Requirement 45 and updates Requirement 39 to “one chat per specification; multiple threads per chat”.
• Adds Decisions 153–154 and Assumption A94 to capture the chosen thread model and the “flat threads + invoked_in_turn_id” stance.
• Updates I111 and the lexicon to reflect chat as a pure container and thread as the conversational-scope primitive.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. No suggestions at this time.

Comment augment review to trigger a new review at any time.

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 5a9ef97. Configure here.}

[cursor]

Race condition allows duplicate thread creation on rapid clicks

Medium Severity

The openFor callback checks the query cache synchronously for an existing open thread, then fires an async fetch to create one if none is found. There is no in-flight guard, so rapid clicks (or multiple calls before the first fetch resolves and invalidateQueries propagates) will each see no existing thread in the stale cache and each create a new one. The old SideChatHost used activeRef and session tracking to prevent this. The new code has no debounce, no optimistic local state, and no ref-based in-flight flag, so duplicate side threads for the same item can be created.

 

^{Reviewed by Cursor Bugbot for commit 5a9ef97. Configure here.}

[cursor]

Focus signal uses item ID, expanding multiple collapsibles

Low Severity

focusedThreadItemId is set to item.id (the knowledge-item ID), not the thread's own ID. In ThreadCollapsible, the focus effect matches on targetItemId. If multiple threads share the same target_item_id (e.g. a closed side thread and a newly opened one for the same item), all matching ThreadCollapsible instances will call setIsExpanded(true) in the same React batch before clearFocusedThread propagates, causing all of them to expand instead of only the intended one.

Additional Locations (1)

• src/client/components/thread-collapsible.tsx#L273-L287

 

^{Reviewed by Cursor Bugbot for commit 5a9ef97. Configure here.}

[kostandinang]

Closing without merge. Substrate decision reversed per #139: schema-level thread is deferred and sub-chats will use the existing chat substrate instead. The Linear issue this PR resolved (FE-710) is cancelled and superseded by FE-716.

This PR is retained as historical reference — the UX work (inline collapsibles, popover retirement, design brief, kickoff turn, streaming, mode toggle, patch staging) will be harvested into the new branch on top of #139 with the corrected substrate.