Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
C++
T11 · Chat: remaining findings
Phase 3 · 38 findings (0 high, 20 medium, 18 low) · Depends on: T01, T05, T06, T07
Goal
Fix the remaining Chat findings after the shared fixes have landed. Skip any finding that a phase 1 task already resolved.
Files to change
These are the likely files. Read them before editing, and change others only if the fix needs it.
src/screens/Chat.tsxsrc/components/ChatToolbar.tsxsrc/components/ComposerControls.tsxsrc/components/ArtifactCard.tsxsrc/components/ChatMedia.tsxsrc/components/SubagentRow.tsxsrc/components/Markdown.tsx
Findings to fix (high and medium)
| # | Severity | State | Issue | Suggested fix |
|---|---|---|---|---|
| 1 | Medium | chat-empty |
"0 subagents ▾" dropdown with zero items | hide at 0 |
| 2 | Medium | chat-empty |
Disabled Send = pale green on pale green, reads as selected | neutral disabled style |
| 3 | Medium | chat-generating |
Header title clipped mid-word ("…short paragrap") with no ellipsis | text-overflow: ellipsis + title attr |
| 4 | Medium | chat-model-picker |
"Running" rendered as a disabled outlined button | status chip (green dot + text), not a button |
| 5 | Medium | chat-template |
Template cards carry a ▾ chevron, implying expand, but the action is "pick" | remove chevron; whole card selectable, "Use template" on hover |
| 6 | Medium | chat-session-log |
Dialog for an unbuilt feature ("will be connected in a future build"), only "Done" | hide the entry until it works |
| 7 | Medium | chat-slash-commands |
/new and /clear have identical descriptions | merge, or /clear = "Clear this conversation" |
| 8 | Medium | chat-slash-commands |
/time-skip "Try the two-stream UI demo · add a prompt" — internal jargon | "Try two versions and keep the better one" |
| 9 | Medium | chat-subagents-statuses |
Failed subagent ("Preview server did not respond") offers no action | Retry / View log on failed rows |
| 10 | Medium | chat-context-length |
Context limit shows 128K; model supports up to 1M (Engine setting) | read limit from engine config |
| 11 | Medium | chat-approval-collapsed |
"▶ Read project notes · done" uses native ▶ and bold H3 weight — louder than the approval card it sits above | small grey row with chevron icon, like tool calls elsewhere |
| 12 | Medium | chat-approval-approved |
Approved change collapses to "· done"; denied says "· denied" — asymmetric, and "done" hides that a write happened | "Approved" (green check) / "Denied" (grey x) |
| 13 | Medium | chat-artifact-cards |
File cards: subtitle "Website · Open in sidebar" reads like a link but isn't; chevron opens a menu — primary click unclear | whole card opens in sidebar; ••• menu for the rest |
| 14 | Medium | chat-message-actions |
Hover "branch" icon button sits on top of the user bubble and clips its text ("…a conversat") | place actions below the bubble with the other icons |
| 15 | Medium | chat-edit-branch |
A new branch shows the empty-state hero + suggestions while the composer holds the edited message | hide empty state in branches; show "Editing from: …" note above composer |
| 16 | Medium | chat-add-menu |
+ menu mixes demo entries ("Try preview examples", "Try chat showcase") with real actions; no way to attach a file to the message | Attach file / Open website; move demos to Help |
| 17 | Medium | chat-tool-access |
Read only / Workspace write / Full access with no description; Full access not marked as risky | one-line description each; amber for Full access |
| 18 | Medium | chat-no-conversations |
Two different empty states: this one (icon, small heading, "Ask a question, work with text…", extra New chat button) vs chat-empty (big heading, suggestions, different subline) | one empty state; drop the redundant New chat button (composer is already live) |
| 19 | Medium | chat-response-complete |
Markdown not rendered: "where" shows literal asterisks | fix inline emphasis parsing |
| 20 | Medium | chat-response-complete |
After a reply, stats show "60 tokens/s · — total tokens · 1%" — total missing | populate total; label 1% ("1% of context") |
Optional polish (low)
Do these only if the task is otherwise done and the budget allows.
| # | Severity | State | Issue | Suggested fix |
|---|---|---|---|---|
| 1 | Low | chat-empty |
3 suggestion cards in 2-col grid leave an orphan | 3-col or 1 row of 3 |
| 2 | Low | chat-generating |
Two chats both titled "Chat showcase · code, to…" in list | dedupe titles / show distinguishing snippet |
| 3 | Low | chat-generating |
Unexplained green dot on "Build a calm website…" list item | legend/tooltip ("Unread" / "Running task") |
| 4 | Low | chat-template |
Templates duplicate but differ from empty-state suggestions (two sets of starters) | single set of starters |
| 5 | Low | chat-slash-commands |
/setup "choose model", /model "Open Models" — plural/multi-model language | "/model Open Model" |
| 6 | Low | chat-slash-commands |
Send turns active with just "/" typed | keep disabled while the input is only a command prefix |
| 7 | Low | chat-open-website-invalid |
"Open preview" same pale-green style whether valid or not | disabled style until valid |
| 8 | Low | chat-showcase-code |
Text "More" under assistant message next to icon actions | ••• icon button |
| 9 | Low | chat-session-statistics |
"LLM time" jargon; "Turns / tool calls 1 / 2" slash-pair | "Thinking & writing"; separate rows |
| 10 | Low | chat-context-length |
Footnote about compaction "not connected in this build" | remove until built |
| 11 | Low | chat-approval-collapsed |
Expand chevron floats mid-row, detached from title and from actions | put chevron at far right or next to title |
| 12 | Low | chat-approval-approved |
Section heading "A change waiting for you" stays after it's resolved | update heading or treat as static demo copy |
| 13 | Low | chat-gif-paused |
With preview open, chat column narrows and the composer wraps the model chip onto its own row; visible scrollbar between panes | collapse composer controls into icons at narrow widths |
| 14 | Low | chat-artifact-cards |
Blockquote "A calmer place…" is indented with no quote bar, reads as misalignment | 3px left rule + muted text |
| 15 | Low | chat-artifact-actions |
Menu says "Show in file explorer" while copy above says "Finder" | OS-aware label |
| 16 | Low | chat-message-actions |
"‹ Latest" pill doesn't say where you are | "Branch 1 of 2 · Back to latest" |
| 17 | Low | chat-edit-branch |
Title gets "(branch)" appended in parentheses | branch icon chip after title |
| 18 | Low | chat-thinking-menu |
Auto/Low/Medium/High with no hint of trade-off | one-line hint each ("Faster" … "Most careful") |
How to check
Run these and fix anything that fails:
npm run typechecknode --test tests/(all existing tests stay green)npm run check(typecheck, UI audit, contrast, measure, luminance)npm run build
Task-specific checks:
- Each finding above is resolved in the running app, in light and dark mode.
- Each finding above is resolved, or marked "already fixed by Txx" or "not reproducible" with a one-line reason.
Screens to re-capture and compare with the audit screenshots in docs/evidence/audit-2026-10-07/:
chat-empty-light.pngchat-generating-light.pngchat-model-picker-light.pngchat-template-light.pngchat-session-log-light.pngchat-slash-commands-light.pngchat-open-website-invalid-light.pngchat-showcase-code-light.pngchat-subagents-statuses-light.pngchat-session-statistics-light.pngchat-context-length-light.pngchat-approval-collapsed-light.pngchat-approval-approved-light.pngchat-gif-paused-light.pngchat-artifact-cards-light.pngchat-artifact-actions-light.pngchat-message-actions-light.pngchat-edit-branch-light.pngchat-add-menu-light.pngchat-thinking-menu-light.pngchat-tool-access-light.pngchat-no-conversations-light.pngchat-response-complete-light.png
Done when
- All checks pass.
- Commit on branch
fix/t11-chatwith messageT11: Chat: remaining findings. - Append a short entry to
audit-fixes/PROGRESS.md: what changed, which findings were fixed, skipped or not reproducible, and anything the reviewer should look at.