[kiwidream]

	Files	Added	Deleted	Net
Test	0	0	0	0
Prod	6	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​62	$\color{#cf222e}{\Huge{\mathbf{−}}}$​4	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​58

ELI5

Dragging the left sidebar narrower left a black strip between the sidebar and the terminal, and the terminal didn't grow into the space. Now the terminal fills it. When the UI is zoomed far out, the sidebar also stops shrinking before its buttons would overlap.

What Changed

Before: on macOS (and paired-web) chrome, the titlebar header above the sidebar set the left column's width by its full content width. The header holds the traffic-light pad, the fork's long Forca 1.4.214 (SwapLabs/<rev>) app name, the sidebar toggle and the history buttons, which add up to about 436px. Dragging the sidebar below that left the column at 436px: the leftover strip showed the transparent app background (black), and the terminal kept its old width.

4ee7b3b: .titlebar-left:not(.titlebar-left-floating) now has contain: inline-size, so only the sidebar sizes the column. min-w-0 on the control group lets the app name clip before any button.

3d193c9: at UI zoom −2.5 and below, the traffic-light pad (80px / zoom) no longer fit beside the buttons in a 220px sidebar, so the toggle overlapped Go back and ignored clicks. The titlebar's existing ResizeObserver now also measures the header's non-shrinkable width: every control, plus the zoom-scaled pad, minus the app name's text. The sidebar uses max(220, that width) as both its drag minimum and its rendered floor. The stored width is not rewritten, so zooming back in restores it.

Why

Containment is the smallest change that stops header content from sizing the column. The alternatives each fall short:

• shortening or hiding the label leaves the zoomed-out case broken;
• an explicit column width would duplicate the sidebar-width state.

Deriving the minimum from the measured controls keeps every button usable at any zoom without letting the header widen the column again.

Linked Issue

N/A: the operator reported it directly. Fork-only (LOCAL:) for now. Upstream has a latent version of the same bug: with the shorter Orca label the gap is 2px at zoom 1 and 14px at zoom −3. Upstreaming is a separate decision.

Visual Proof

Measured in the hidden e2e app (Electron on Linux with a macOS user agent), in dark and light themes. The terminal's column count at sidebar widths 480 → 360 → 280 → 220 (a drag to 150 stops at the 220 minimum):

	480	360	280	220
Before (column stuck at 436px)	66	71	71	71
After	66	80	89	97

Before 3d193c9, at UI zoom −2.5 and −3 with the sidebar at its minimum, the toggle overlapped Go back by 11px and 22px. After it:

• the effective minimum is 241px at −2.5 and 253px at −3, and stays 220px at every other zoom tested;
• there are no overlaps, and a real click on the toggle collapses the sidebar at every minimum;
• zoom 0 geometry is identical to 4ee7b3b;
• Linux and Windows custom chrome are unchanged (minimum 220px at every zoom).

Before/after screenshots exist for every width, zoom and chrome (on minidev-alex1) but are not attached here, because no image-upload tool is available on the VM. I can add them on request.

Testing

• An e2e repro spec, not committed because it needs a user-agent spoof, covered:

  • widths 480 to 150;
  • zoom −3, −2.5, −1.5, 0 and +2;
  • Mac, Linux and Windows chrome;
  • collapsed/floating mode, full screen, the app name hidden, and the right sidebar open and closed.

• An independent adversarial verification run confirmed 4ee7b3b and found the zoom-out overlap that 3d193c9 fixes. 3d193c9 was then measured across the same matrix.

• pnpm check:changed --maxWorkers=2 --base=refs/remotes/swaplabs/swaplabs/main passed (20/20) on both commits.

• Not tested on a physical Mac; the macOS chrome was exercised through the user-agent spoof.

• I manually tested these changes locally (hidden Electron renderer on Linux, Mac chrome via user-agent spoof)

• Automated tests added/updated, or explained why not below: the geometry depends on macOS chrome, which the e2e suite only renders under a user-agent spoof, so no committed test was added.

AI Disclosure

Prepared by Claude Code (Claude Opus 5.5) at the operator's request.

Review

Local /code-review medium before the first push raised 8 candidate findings. None blocks: the measured runs above show no overlap or gap at any tested width, zoom or chrome. Deferred, with reasons:

• At UI zoom −2.5 and below, the measured minimum includes the history buttons, which mount only in some views. Switching to a view without them lowers the minimum by about 60px, so the sidebar can change width. Below −2.5 only.
• The ResizeObserver subscribes to the row's grandchildren once per effect run. A remount without a dependency change, followed by a zoom change, could leave the minimum stale. Speculative.
• Each observer callback now reads styles for about 8 elements, including on every drag frame. The reads run after layout and an unchanged value does not re-render, so the cost is minor. It could short-circuit.
• Not adopted: deriving the minimum from the zoom factor alone (220 + 80/z − 80). That over-reserves by about 25px and would raise the minimum at moderate zoom-outs, such as −1.5, where it isn't needed.
• By design: a drag that starts at the raised width saves the dragged width, as any drag does.
• Not acted on: two speculative cases, the sidebar's error fallback and a few-px margin under-count.

Agent skill upstream boundary

• Not applicable: renderer layout only; no agent skill.

Notes

• Security: none affected.
• Cross-platform: Linux and Windows custom chrome measured unchanged.
• Remote and SSH: renderer-only.
• Mobile: has no sidebar.
• Backward compatibility: the stored sidebar width is never rewritten.
• Performance: one more element measured by the existing ResizeObserver.

Checklist

• This PR is small and focused
• I explained what changed and why (ELI5, the user-facing before/after, the mechanism, and why over the alternatives)
• Before/after screenshots or videos attached for UI changes, or N/A with reason: measured numbers above; image files available on request
• Self-reviewed for correctness, security, and performance
• Cross-platform, SSH/remote, and path/shortcut impact considered (or N/A)
• pnpm lint, pnpm typecheck, pnpm test, and pnpm build pass (or CI will cover; local preferred): check:changed (format, lint, typecheck, affected tests) passed; the full build is left to CI

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-28T14:42:48.695868Z	3d193c9	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.