levelcodeai
diff --git a/docs/CHAT-TYPOGRAPHY.md b/docs/CHAT-TYPOGRAPHY.md
@@ -45,12 +45,16 @@ Two observations worth stating plainly, because they explain almost everything:
 
 ### D1 — Constrain the measure. This is the single biggest lever.
 
-`max-width` on the prose column, centred, with the container still full-bleed so cards, code blocks
-and the composer keep their current behaviour.
+`max-width` on the prose column, centred, with the container still full-bleed so cards and code blocks
+keep their current behaviour. *(D9 later brought the composer into the same column; T1 shipped with it
+still full-bleed.)*
 
-**Target 680px** — measured, not estimated. An earlier draft of this section said "~72ch, which lands
-near 640–700px". Both halves were wrong, and the correction is worth keeping because the mistake is
-easy to repeat:
+**Target 820px**. T1 shipped 680px, derived from first principles below. That number was later
+overruled by the reference itself — see the revision at the end of this decision. The derivation is
+kept because its correction is the reusable part; the figure it produced is not.
+
+An earlier draft of this section said "~72ch, which lands near 640–700px". Both halves were wrong, and
+that correction is worth keeping because the mistake is easy to repeat:
 
 | measured in the shipped font at 13px | |
 | --- | --- |
@@ -62,9 +66,21 @@ characters would be a 422px column, narrower than the sidebar. The print-typogra
 characters does not transfer to a technical chat: it assumes prose without identifiers, file paths or
 code, and a 422px column would wrap every code block constantly.
 
-680px is chosen against the measurement: **~116 characters at 13px**, down from 154 at editor width,
-while staying wide enough that a fenced block is still readable. When T2 raises the prose size the
-same cap tightens to ~108 characters, which is the right direction.
+680px was chosen against that measurement: ~116 characters at 13px, down from 154 at editor width,
+while staying wide enough that a fenced block is still readable.
+
+**Revised to 820px after comparing against the reference directly.** Side-by-side at the same window
+width, the Claude Code console's column is **~815px** against our 680 — it uses about half the
+available width where we used 41%. The user's report was "the content space is too narrow", and on a
+design parameter chosen from first principles versus the artefact we are explicitly trying to match,
+the artefact wins.
+
+Stated honestly: 820px is **~130 characters** at the shipped 14px prose size, which is well outside the
+print range this decision already argues does not transfer. Two things make that acceptable rather than
+sloppy — the reference demonstrably reads well at that width, and this column carries tables, file
+paths and fenced code, none of which wrap gracefully at 680. The risk section's "screenshots are not
+measurements" caveat applies: ~815px is read off a screenshot, so 820 is a round number near a
+measured one, not a precise transcription.
 
 The elegant part: **in the sidebar this is a no-op.** The container is already narrower than the cap,
 so nothing moves for existing users. It only takes effect in the editor tab, which is exactly the
@@ -196,6 +212,40 @@ Measured in headless Chrome at a 680px column, against `develop`:
 Tint stays as the secondary cue rather than being removed: side alone would fail on any surface that
 reflows the log to a single column.
 
+### D9 — One column for the whole panel, not just the transcript.
+
+T1 bounded the **transcript** and nothing else. `#composer`, `#status` and the four notice bars are
+siblings of `#log`, not children, so none of them saw the cap: at editor width the input was a
+**~1580px box sitting under an 820px conversation**. Side by side with the reference — where the
+composer sits directly under the text it answers — this was the single most obvious difference left,
+more than any type choice.
+
+**The measure moves from `#log` to `body`.** This is the load-bearing part, not tidying. A custom
+property declared on the log is invisible to the log's siblings, which is precisely how the split
+arose. On `body`, every element in the shell resolves the same value.
+
+Two consequences that are easy to miss:
+
+- **`--shell-x`.** The shell insets by `calc(100% - 2 * var(--shell-x))`, the same value `#log` pads
+  with, rather than a bare `width: 100%`. Without it the edges agree only where the cap binds, and
+  out-dent by the log's padding everywhere else — including every sidebar, which is the width most
+  users are actually in.
+- **The runtime override moves too.** `chat.proseWidth` used to be written onto `#log`. Left there it
+  would resize the conversation and leave the composer on the stylesheet default — reopening this exact
+  split, but only for users who set the setting, which is the worst place for it to hide.
+
+Verified in headless Chrome at four widths. Composer and prose edges, left and right:
+
+| viewport | before (prose / composer) | after |
+| --- | --- | --- |
+| 1600px | 680 / **1584** — 452px out-dent per side | 820 / 820 — **Δ 0.0** |
+| 900px | 680 / 884 | 820 / 820 — **Δ 0.0** |
+| 800px | 680 / 784 | 752 / 752 — **Δ 0.0** |
+| 420px (sidebar) | 476 / 484 | 476 / 476 — **Δ 0.0** |
+
+The 420px row is the argument for `--shell-x` on its own: the sidebar was misaligned by 4px a side
+before this, quietly, in the surface almost everyone uses.
+
 ---
 
 ## 3. Slices
@@ -226,6 +276,11 @@ left and full-measure. **Exit:** a short turn renders as a short right-flush bub
 below the column, the assistant is untouched, and the prose inside the bubble is still left-aligned —
 all four measured, not eyeballed.
 
+**T7 — the shell column** *(S)*. D9, plus D1's revision to 820px. Ships: the measure moves to `body`,
+`--shell-x`, the shell rule over the composer/status/notice bars, and the runtime override retargeted.
+**Exit:** composer and prose edges agree to **0.0px at 1600 / 900 / 800 / 420**, and `chat.proseWidth`
+moves both together — measured, not eyeballed.
+
 **T5 — the escape hatch** *(S)*. D7. **Folded into T2 and shipped with it.** Sequencing it last was a
 mistake: T2 is the one slice that changes what every existing user sees, and shipping a divisive
 change with no way back is worse than not shipping it. The plumbing is also shared — once one custom
@@ -234,10 +289,12 @@ property reaches the webview from settings, the second is a line — so splittin
 Sequencing: T1 first and alone — it may turn out to be most of the perceived fix, and shipping it
 by itself is the cheapest way to find out before spending effort on T2–T4.
 
-That held up: T1–T2, T4 and T6 have shipped in that order, each visible on its own. **T3 is the only
-slice of this plan still outstanding.** D8/T6 was not in the original decomposition — it came from
-looking at the reference again after T4, which is the argument for shipping slices small enough to
-look at.
+That held up: T1–T2, T4, T6 and T7 have shipped in that order, each visible on its own. **T3 is the
+only slice of this plan still outstanding.** Neither D8/T6 nor D9/T7 was in the original decomposition
+— both came from looking at the reference again after shipping, which is the argument for slices small
+enough to look at. D9 in particular was invisible from inside the plan: T1's own wording said the
+composer "keeps its current behaviour", and it took a side-by-side screenshot to notice that was the
+bug rather than the scope.
 
 ---
 

diff --git a/extensions/levelcode-ai/media/chat.html b/extensions/levelcode-ai/media/chat.html
@@ -22,13 +22,16 @@
     color: var(--vscode-foreground);
     background: transparent;
     display: flex; flex-direction: column; height: 100vh;
-  }
 
-  /* ---- conversation log ---- */
-  #log { flex: 1; overflow-y: auto; padding: 12px; display: flex; flex-direction: column; gap: 12px;
-    /* T1 (docs/CHAT-TYPOGRAPHY.md D1) — the reading measure. A custom property so T5 can hand it to a
-       setting later without touching the rules below. */
-    --prose-max: 680px;
+    /* T7 (docs/CHAT-TYPOGRAPHY.md D9) — THE SHELL COLUMN. These three live on `body`, not on `#log`,
+       because the measure is a property of the whole panel: the transcript, the composer and the status
+       row are all children of `body`, and they have to resolve the SAME value or they cannot line up.
+       Declared on `#log` (where T1 put them) only the transcript could see them, which is exactly how
+       the composer ended up spanning the full panel while the prose sat in a 680px column. */
+    --prose-max: 820px;
+    /* The horizontal inset shared by the log's padding and the shell column's width, so the composer's
+       left edge lands ON the prose's left edge at every width rather than only where the cap binds. */
+    --shell-x: 12px;
     /* T2 (D2) — the READING type. `--vscode-font-size` is the size of menu labels and tree rows: right
        for chrome, wrong for three paragraphs of explanation. Message bodies alone get their own size
        and leading; every control around them keeps inheriting the workbench, so the panel still
@@ -39,7 +42,28 @@
        the wrong way. At the default 13px this still resolves to 14px, so nothing moves for anyone who
        has not changed it. */
     --prose-size: calc(var(--vscode-font-size, 13px) + 1px);
-    --prose-leading: 1.65; }
+    --prose-leading: 1.65;
+  }
+  @media (min-width: 760px) { body { --shell-x: 24px; } }
+
+  /* Everything below the log shares the transcript's column: same width, same centre, same edges.
+     Before this each one spanned the whole panel, so at editor width the composer was a 1600px box
+     under a 680px conversation — the single thing that most made the panel look unlike the reference,
+     where the input sits directly under the text it answers.
+     `calc(100% - 2 * --shell-x)` rather than a bare `width: 100%` + margin: without it these boxes
+     would out-dent the transcript by the log's padding in every panel narrower than the cap, which is
+     most sidebars. Overlays are `position: fixed` and deliberately excluded. */
+  #bgTasksBar, #ctxBar, #planBar, #reviewBar, #workbar, #composer, #status {
+    width: calc(100% - 2 * var(--shell-x));
+    max-width: var(--prose-max);
+    margin-inline: auto;
+  }
+
+  /* ---- conversation log ---- */
+  /* The measure, the type and the inset are all declared on `body` (see above) so the composer and the
+     status row resolve the same values the transcript does. `--shell-x` is the log's horizontal padding
+     AND the shell column's inset, which is what keeps the two edges on the same line. */
+  #log { flex: 1; overflow-y: auto; padding: 12px var(--shell-x); display: flex; flex-direction: column; gap: 12px; }
   /* Log is a flex column → children default to flex-shrink:1. Any child with overflow:hidden (e.g. the
      approval/edit cards) then gets an auto min-size of 0 and the flex algorithm crushes it to an invisible
      line once the log overflows. Pin every child to its natural height; the log itself scrolls instead. */
@@ -48,12 +72,15 @@
      in this file are all cards, dialogs and the empty state. In a 380px sidebar the container bounded
      it, so it never looked wrong — but the chat can now open as an editor tab (AI: Open Chat in
      Editor), and at 900px the same CSS produced ~154-character lines.
-     Measured in the shipped font at 13px: the average prose character is 5.86px, so 680px is ~116
-     characters — a real improvement while staying wide enough that a fenced code block does not wrap
-     constantly. NOT expressed in `ch`: `0` is 8.13px here, 39% wider than average text, so a `ch` cap
-     silently overshoots by about a third.
+     The cap itself is `--prose-max`, declared on `body`, and is deliberately NOT restated here. This
+     comment carried its own copy of the number and said 680 for exactly as long as it took the cap to
+     become 820 — which is what a second copy of a value is always for. docs/CHAT-TYPOGRAPHY.md D1 has
+     how it was chosen; webviewCss.test.js fails if this comment starts quoting a length again.
+     NOT expressed in `ch`: `0` measures 8.13px against an average prose character of 5.86px, so a
+     `ch` cap silently overshoots by about a third. Those two figures stay because they are facts about
+     the typeface, not a decision that can be revised.
      Applied to EVERY direct child, not just .msg, so messages, cards and the activity timeline share
-     one column instead of drifting apart at width. Below 680px this is inert — which is why the
+     one column instead of drifting apart at width. Below the cap this is inert — which is why the
      sidebar is untouched. */
   #log > * { width: 100%; max-width: var(--prose-max); margin-inline: auto; box-sizing: border-box; }
   /* ---- per-response copy button (hover-reveal under a completed assistant message) ---- */
@@ -174,7 +201,7 @@
      opens the editor tab sees nothing move. `min-width` here is the WEBVIEW's own width — the iframe
      is sized to the panel, which is why the existing #empty/dialog rules already use vw/vh. */
   @media (min-width: 760px) {
-    #log { padding: 20px 24px; gap: 1.15em; }
+    #log { padding: 20px var(--shell-x); gap: 1.15em; }
     .msg .body p { margin-bottom: 1em; }
     .msg .body ul, .msg .body ol { margin-bottom: 1em; }
     /* More space ABOVE a heading than below it, so it binds to the section it introduces rather than
@@ -745,7 +772,7 @@
 
   /* ---- composer ---- */
   #composer {
-    margin: 8px; padding: 8px 8px 6px;
+    margin: 8px auto; padding: 8px 8px 6px;
     border: 1px solid var(--border); border-radius: var(--radius);
     background: var(--field-bg);
     display: flex; flex-direction: column; gap: 7px;
@@ -3573,10 +3600,10 @@
     if (m.type === 'config'){
       // D7 — the escape hatch. 0 means "follow the default": clearing the property lets the stylesheet
       // value win again, rather than pinning it to whatever the default happened to be at the time.
-      const rootLog = document.getElementById('log');
-      if (rootLog){
-        rootLog.style.setProperty('--prose-size', m.proseSize ? m.proseSize + 'px' : '');
-        rootLog.style.setProperty('--prose-max', m.proseWidth ? m.proseWidth + 'px' : '');
+      const shell = document.body;
+      if (shell){
+        shell.style.setProperty('--prose-size', m.proseSize ? m.proseSize + 'px' : '');
+        shell.style.setProperty('--prose-max', m.proseWidth ? m.proseWidth + 'px' : '');
       }
       const label = m.provider === 'claude' ? prettyModel(m.model) : m.model;
       const mEl = document.getElementById('model');