Keep at least two shells visible in the Terminal rail, and let its sections resize

[jonathanKingston]

Problem

The Terminal tab's left rail stacks Shells, Agent tasks, Background tasks and Ports as fixed-shrink blocks whose percentage max-heights never resolve. A long agent-task list therefore pushes the Shells list down to a sliver, so the open terminals become invisible while the rail fills with task rows.

Change

• Every rail section is now a terminal-rail-section that shares the rail's height (flex: 1 1 33%), and the rail scrolls instead of crushing sections when their minimums exceed its height.
• New terminal-rail-resizer.ts gives each visible section a minimum of its header plus two rows, and holds Shells to at least a third of the rail, so at least two terminals stay visible however many tasks or ports are listed.
• Divider handles between visible sections drag (or ArrowUp/ArrowDown when focused) to rebalance adjacent sections without crossing either minimum. Handles hide when a neighbour is hidden and carry an aria-label naming the two sections.

This ports work that was left uncommitted in a worktree on the old lineage (copse/for-shells-can-we-make-the-panel-sections-5d38eb); it never reached a PR.

Verification

• terminal-rail-resizer.test.ts covers the minimums, handle visibility and pointer/keyboard resizing; pnpm test -- terminal-rail-resizer agent-tasks ports-section supervised-tasks passes (30 tests).
• pnpm run typecheck, pnpm run lint and pnpm run format:check are clean.
• tests/e2e/supervised-tasks.e2e.ts now seeds ports, drags both handles and asserts every section holds its minimum. It could not be run on my machine: the unmodified spec from main fails identically in its before all hook (the known local reloadSession timeout), so CI is the check for it.
• Visual probe with the built app.css and the compiled resizer, 3 shells + 15 agent tasks in a 900px rail: Shells held 300px with all 3 shells visible, agent tasks scrolled within their section; dragging the Shells handle up stopped at the Shells minimum, dragging down stopped at Agent tasks' two-row minimum.

🤖 Generated with Claude Code

[github-actions]

🖥️ PR preview

• Marketing site
• Browser demo
  Interactive, with a mocked backend — jump to a scenario:
• Builds a cupcake site
• Agent-proposed edits open the Changes panel
• Markdown list indentation
• Responsive composer footer
• Subagent display visual reference
• Subagent stays ungrouped beside reading tools
• Semantic search subagent markdown
• Light-theme approval with a bright accent
• Grouped outside-sandbox command approval
• Remote desktop discovered-port list with one selected
• Settings scroll + sticky footer geometry
• Per-model reasoning / temperature / top-p
• Thread switch hydration notice
• Thread switch hydration failure
• Chat layout styling
  _{Static demo of the renderer against a mock API — the browser pane, terminal, and all main-process IPC are stubbed, so it never proves main-process behaviour. Rebuilt whenever a push changes the demo bundle; these links go live once the queued deploy lands, shortly after this comment. Removed when the PR closes.}

[jonathanKingston]

Review: terminal rail resizer

Verdict: changes needed. The resizer hangs the renderer on boot, which is why all 16 app-loading demo specs in the build job timed out. Small fix, reproduced end to end.

Blocking: infinite MutationObserver loop

src/renderer/views/terminal-rail-resizer.ts:161-165 and :175-180. sync() assigns handle.hidden = section.hidden || before === null on every call, and the MutationObserver on root (subtree: true, attributeFilter: ['hidden']) sees the handles because they are children of root. Per the DOM spec, setting hidden to true when it is already true still queues a mutation record ("change an attribute" always enqueues), while hidden = false on an absent attribute does not. So as soon as any section is hidden at mount, sync reschedules itself forever on the microtask queue. In production that is every boot: mountSupervisedTasks (supervised-tasks.ts:17-20) and mountPortsSection (ports-section.ts:66-69) both create their section with hidden: true, and Agent tasks hides itself when empty.

Evidence:

• Minimal page with the compiled resizer and two hidden sections: headless Chromium never finishes and is killed by the timeout. The same page with no hidden sections finishes instantly.
• PR demo build (pnpm run build && pnpm run build:demo, served from dist/demo): /?scenario=chat-layout-styling and /?scenario=settings-footer both time out on page.goto at 20 s and on page.evaluate('1+1') at 10 s. With the guard below and only build:demo rerun, the same scenarios load in about 5 s with 4 sections, 3 handles, 2 hidden handles, and the prompt input present.
• A spy observer counting hidden records after mount reports 52 (guard tripped) on the PR code and 2 with the fix.

Minimal fix:

const hidden = section.hidden || before === null
if (handle.hidden !== hidden) handle.hidden = hidden

Observing the sections individually instead of root with subtree would also work; the guard is smaller and cheaper.

Please also add a regression test. The current tests miss this because records are delivered on a microtask and the first test calls destroy() synchronously, disconnecting the observer before delivery; the second test has no hidden section. happy-dom does deliver same-value records (probe: two same-value writes produced 2 records), so a test that mounts with a hidden section, awaits a setTimeout, and counts hidden mutation records via a spy observer with a guard that calls destroy() above about 50 records (so it fails instead of hanging the runner) catches it: 52 today, 2 with the fix.

Validation

Check	Result
pnpm test -- terminal-rail-resizer agent-tasks ports-section supervised-tasks	31/31 pass
pnpm run build && pnpm run build:demo	pass
Demo scenarios on the PR build	hang
Demo scenarios on the patched build	responsive
Dry merge with origin/main (2 behind)	clean

Non-blocking

• Screenshot evidence. tests/e2e/supervised-tasks.e2e.ts calls saveAppScreenshot('supervised-tasks-waiting.png') after the two drags, so the reference will change, but no PNG is in the diff because e2e never ran. Commit the candidate from the reference-screenshot-candidates-<run-id> artifact once CI is green. The spec logic itself looks right: __copseE2e.setPortRows exists behind COPSE_E2E=1 (preload/index.ts:1306, register-handlers.ts:2615), Agent tasks is hidden with no tasks so "Resize Shells and Background tasks" is the correct label, and dragAndDrop({x, y}) has precedent in monaco-selection-chat.e2e.ts:48.
• sectionMinimum writes 8px minimums while #terminals-list-host is still hidden (index.html:63; all rects are 0). The ResizeObserver repairs this when the host is shown, but an early return on root.clientHeight === 0 would be cleaner.
• The focusable role="separator" should expose aria-valuenow, aria-valuemin and aria-valuemax. The aria-label is also left stale when before is null, harmless because the handle is hidden then.
• After a drag, visible sections carry flex: <px> 1 0px while a section shown later keeps the stylesheet flex: 1 1 33.333%. It works, but consider resetting style.flex when a section transitions hidden to visible.
• Empty sections get a header-plus-padding minimum since measuredRowHeight (:17-22) falls back to a missing first child. Fine given empty sections are hidden and Shells has the one-third floor; a comment would help.

ResizeObserver feedback checked and not a loop: the host is flex: 1; min-height: 0; overflow-y: auto, so child min-height writes change scrollHeight only, and sync is idempotent once a scrollbar appears.

---

Generated by Claude Code

[jonathanKingston]

Pushed 6245525 with the fix from the review: sync() now only writes handle.hidden when the value changes, so the MutationObserver no longer re-triggers itself; syncMinimums skips writing minimums while the host is hidden; plus a regression test that mounts with hidden sections and counts hidden mutation records (52 before the fix, 2 after). Verified on the demo build: page.evaluate answers in 9 ms and the rail shows 3 handles with 2 hidden. Typecheck, eslint, oxfmt, dead-code and the four rail suites (31/31) pass. Reference PNGs are left to CI's candidate artifact.

---

Generated by Claude Code

[copse-release-bot]

Reference screenshots

CI run 33917334411 published no changed screenshot candidates for 9ce519005bc3.
Any review PR for an older source SHA has been closed.