fix(test): resolve shell/cwd unit-test failures surfaced by GitHub-hosted runners (LAC-2386)

[lacymorrow]

Summary

Fixes the 19 unit (linux) failures that surfaced on PR #28 when CI moved to GitHub-hosted runners (run 27512032937). These were real pre-existing failures, invisible because Blacksmith runs on dev had been cancelled for weeks.

Stacked on #28 (LAC-2385/blacksmith-to-github-runners) so the unit jobs here actually run on GitHub-hosted runners. Merging this turns #28 green, which then merges to dev.

Root causes (three, all environment-sensitive)

1. getCwd() degraded to process.cwd() when the ambient (AsyncLocalStorage) instance context wasn't entered. Only CLI bootstrap() enters it; Effect-based callers run with InstanceRef instead. On CI the process cwd is the runner checkout dir, so shell permission patterns and cwd resolution pointed at the wrong directory — the permission ask hook never fired (requests.length 0 vs 1). Fix: shell tool / session prompt / system prompt pass the instance directory as an explicit fallback, and the test fixture now enters the ambient instance context the way production bootstrap() does.

2. InstanceRef lost in lash session tests. Upstream 0c9cfe9 (refactor(instance): remove legacy runtime fallback) removed the Instance.current fallback from attach(), so tests calling AppRuntime.runPromise from plain async code died with InstanceRef not provided. New test/fixture/app.ts#runTestApp provides InstanceRef explicitly — upstream's intended pattern (Effect.provideService(InstanceRef, ctx)).

3. cwd-sentinel cleanup ate real output. The shell wrapper echoes a __LASH_CWD__ sentinel after each command. Cleanup truncated everything from the start of the sentinel's line, which (a) dropped unterminated stdout sharing that line (printf out), and (b) dropped stderr chunks that arrive after the sentinel in the merged stdout+stderr stream. stripSentinel() now splices out only the sentinel line.

Also resets the shell-mode cwd singleton between tmpdir fixtures so a cd in one test can't leak into the next.

Verification

• bun test test/session/prompt*.test.ts test/tool/shell.test.ts test/plugin/cwd.test.ts — 95 pass / 0 fail (was 19 fail in CI, reproduced locally)
• bun run typecheck clean
• Full package suite run locally; unit (linux) on this PR is the authoritative gate

Out of scope

unit (windows) has 112 pre-existing failures — separate follow-up issue (LAC-2386 acceptance is linux green).

Closes LAC-2386.

[gemini-code-assist]

Code Review

This pull request introduces significant architectural changes, including a new layout design system, a refactored server management controller, and updated performance diagnostics. It also migrates several components to a V2 UI library and improves the robustness of session tab management. I have identified several critical and high-severity issues, primarily related to parameter shadowing in the tab navigation component, potential null pointer dereferences in draft promotion, and missing imports in error handling blocks. Please address these bugs before merging.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Critical Reactivity Bug: Parameter Shadowing and Accessor Invocation

Inside <Show when={props.session()}>, the render function parameter is named session (line 192). This shadows the accessor function props.session (or any helper named session).

Because session is the parameter representing the resolved Session object (not a function), calling session() as a function on lines 214, 215, 222, and 242 will throw a TypeError: session is not a function at runtime, causing a complete crash of the tab navigation.

To fix this, we should simply access the properties directly on the session parameter (e.g., session.directory, session.id, session.title) without invoking it as a function.

      <Show when={props.session()}>
        {(session) => {
          return (
            <a
              data-slot="tab-link"
              data-titlebar-tab-link
              href={props.href}
              draggable={false}
              onDragStart={(event) => {
                event.preventDefault()
                event.stopPropagation()
              }}
              onClick={(event) => {
                event.preventDefault()
                if (editing()) return
                if (props.suppressNavigation?.()) return
                props.onNavigate()
              }}
              class="flex h-full min-w-0 flex-1 flex-row items-center gap-1.5 text-[13px] font-medium text-v2-text-text-faint group-data-[active='true']:text-v2-text-text-base group-data-[editing='true']:text-v2-text-text-base [-webkit-user-drag:none]"
            >
              <span data-slot="project-avatar-slot">
                <SessionTabAvatar
                  project={project()}
                  directory={session.directory}
                  sessionId={session.id}
                  activeServer={props.activeServer}
                />
              </span>
              <span
                ref={(el) => {
                  titleEl = el
                  titleEl.textContent = session.title
                }}
                data-slot="tab-title"
                data-titlebar-tab-title
                class="min-w-0 flex-1 outline-none leading-4"
                classList={{
                  "overflow-hidden text-clip whitespace-nowrap": !editing(),
                  "select-text": editing(),
                }}
                contenteditable={editing() ? true : undefined}
                onDblClick={openRename}
                onKeyDown={(event) => {
                  event.stopPropagation()
                  if (event.key === "Enter") {
                    event.preventDefault()
                    void closeRename(true)
                    return
                  }
                  if (event.key !== "Escape") return
                  event.preventDefault()
                  titleEl.textContent = session.title
                  void closeRename(false)
                }}
                onBlur={() => void closeRename(true)}
                onPointerDown={(event) => {
                  if (!editing()) return
                  event.stopPropagation()
                }}
                onClick={(event) => {
                  if (!editing()) return
                  event.preventDefault()}
                }
              />
            </a>
          )
        }}
      </Show>

[lacymorrow]

Thanks — this file is not authored in this PR. It arrived via the upstream sync (sst/opencode dev, merged to our dev in #31) and only appears in this diff because the base branch (LAC-2385/blacksmith-to-github-runners, PR #28) predates that sync merge. This PR only changes unit-test files (packages/opencode/test/**) to fix CI failures on GitHub-hosted runners (LAC-2386). Upstream code concerns are tracked against upstream, not patched in a test-fix PR — patching them here would create merge conflicts on the next sync.

[gemini-code-assist]

High Severity: Potential Null Pointer Dereference

tabs.draft(draftID) can return undefined if the draft is not found or has been cleared. Accessing .server on undefined will throw a TypeError at runtime. We should safely retrieve the draft first before promoting it.

        const draftID = search.draftId
        const draft = draftID ? tabs.draft(draftID) : undefined
        if (draftID && draft) {
          tabs.promoteDraft(draftID, { server: draft.server, sessionId: session.id })
        } else {
          navigate(`/${base64Encode(sessionDirectory)}/session/${session.id}`)
        }

[lacymorrow]

Thanks — this file is not authored in this PR. It arrived via the upstream sync (sst/opencode dev, merged to our dev in #31) and only appears in this diff because the base branch (LAC-2385/blacksmith-to-github-runners, PR #28) predates that sync merge. This PR only changes unit-test files (packages/opencode/test/**) to fix CI failures on GitHub-hosted runners (LAC-2386). Upstream code concerns are tracked against upstream, not patched in a test-fix PR — patching them here would create merge conflicts on the next sync.

[gemini-code-assist]

High Severity: Circular Dependency & Logic Bug

When checked is false (meaning the user turned OFF the new layout designs), they are switching back to the legacy layout and should be shown the legacy settings dialog (@/components/dialog-settings).

However, the code currently imports @/components/settings-v2 (which is the V2 settings dialog) and shows DialogSettings from V2 again, creating a circular dependency and a broken user experience. We should import @/components/dialog-settings instead.

                void import("@/components/dialog-settings").then((module) => {
                  dialog.show(() => <module.DialogSettings />)
                })

[lacymorrow]

Thanks — this file is not authored in this PR. It arrived via the upstream sync (sst/opencode dev, merged to our dev in #31) and only appears in this diff because the base branch (LAC-2385/blacksmith-to-github-runners, PR #28) predates that sync merge. This PR only changes unit-test files (packages/opencode/test/**) to fix CI failures on GitHub-hosted runners (LAC-2386). Upstream code concerns are tracked against upstream, not patched in a test-fix PR — patching them here would create merge conflicts on the next sync.

[gemini-code-assist]

High Severity: Missing Import / ReferenceError

showRequestError is called in the catch block of handleRemove, but it is not imported in this file. This will cause a runtime crash if an error occurs. We should import showRequestError or use showToast directly.

[lacymorrow]

Thanks — this file is not authored in this PR. It arrived via the upstream sync (sst/opencode dev, merged to our dev in #31) and only appears in this diff because the base branch (LAC-2385/blacksmith-to-github-runners, PR #28) predates that sync merge. This PR only changes unit-test files (packages/opencode/test/**) to fix CI failures on GitHub-hosted runners (LAC-2386). Upstream code concerns are tracked against upstream, not patched in a test-fix PR — patching them here would create merge conflicts on the next sync.

[gemini-code-assist]

Medium Severity: Potential TypeError on Undefined Resource

wslServers() is returned by useFilteredWslServers and can be undefined during loading or if the resource hasn't resolved yet. Accessing .length directly on undefined will throw a TypeError. We should use optional chaining wslServers()?.length to make this safe.

Suggested change

	() => controller.sortedItems().filter((item) => !isWslServer(item)).length + wslServers().length > 1,
	() => controller.sortedItems().filter((item) => !isWslServer(item)).length + (wslServers()?.length ?? 0) > 1,

[lacymorrow]

Thanks — this file is not authored in this PR. It arrived via the upstream sync (sst/opencode dev, merged to our dev in #31) and only appears in this diff because the base branch (LAC-2385/blacksmith-to-github-runners, PR #28) predates that sync merge. This PR only changes unit-test files (packages/opencode/test/**) to fix CI failures on GitHub-hosted runners (LAC-2386). Upstream code concerns are tracked against upstream, not patched in a test-fix PR — patching them here would create merge conflicts on the next sync.

[gemini-code-assist]

Medium Severity: Potential TypeError on Undefined Resource

wslServers() can be undefined during loading. Accessing .length directly on undefined will throw a TypeError. We should use optional chaining wslServers()?.length to make this safe.

Suggested change

	when={filtered().length > 0 || wslServers().length > 0}
	when={filtered().length > 0 || (wslServers()?.length ?? 0) > 0}

[lacymorrow]

Thanks — this file is not authored in this PR. It arrived via the upstream sync (sst/opencode dev, merged to our dev in #31) and only appears in this diff because the base branch (LAC-2385/blacksmith-to-github-runners, PR #28) predates that sync merge. This PR only changes unit-test files (packages/opencode/test/**) to fix CI failures on GitHub-hosted runners (LAC-2386). Upstream code concerns are tracked against upstream, not patched in a test-fix PR — patching them here would create merge conflicts on the next sync.

[gemini-code-assist]

Medium Severity: Commented-out Code

The persisted wrapper is commented out, meaning the dismissed state of the help button is not persisted across page reloads. If this is intentional, please clean up the commented-out code to maintain readability. Otherwise, uncomment it to enable persistence.

Suggested change

	const [state, setState] = /* persisted(Persist.global("help-button"), */ createStore({ dismissed: false }) /* ) */
	const [state, setState] = createStore({ dismissed: false })

[lacymorrow]

Thanks — this file is not authored in this PR. It arrived via the upstream sync (sst/opencode dev, merged to our dev in #31) and only appears in this diff because the base branch (LAC-2385/blacksmith-to-github-runners, PR #28) predates that sync merge. This PR only changes unit-test files (packages/opencode/test/**) to fix CI failures on GitHub-hosted runners (LAC-2386). Upstream code concerns are tracked against upstream, not patched in a test-fix PR — patching them here would create merge conflicts on the next sync.