feat(web,hub,cli): show machine health in session sidebar

[heavygee]

Summary

• CLI: collect OS metrics (cpuPercent, memoryPercent, load1m on Unix, cpuCount) and attach optional health to existing machine-alive keepalives (~20s).
• Hub: persist health on machine cache entries and rebroadcast when values change.
• Web: session sidebar machine row becomes a compact bordered tile (name, OS, session count) with inline CPU/RAM meters; tooltip shows capacity status, per-metric bars, load on Linux, and "CPU across all N cores" when cpuCount is known.

Test plan

• bun typecheck
• bun run test — 1121/1122 pass; one pre-existing CLI runner integration version-mismatch test flakes unrelated to this diff
• Hub alive handler tests for health payload
• Web presentation + component tests
• Dogfood on operator driver soup (:3006) with Linux + Windows runners reporting health

Fixes #961

[github-actions]

Findings

• [Major] Delayed heartbeat timer survives stop/shutdown — see inline comment at cli/src/api/apiMachine.ts:502.

Summary

• Review mode: initial
• One lifecycle issue found in the new machine health heartbeat path. It can leave a heartbeat interval running after disconnect/shutdown if the socket stops during the new 50 ms priming delay.

Testing

• Not run (automation). Suggested coverage: fake-timer test that calls startKeepAlive(), stops/shuts down before 50 ms, advances timers, and asserts no machine-alive emit or interval remains.

HAPI Bot

[heavygee]

Addressed the [Major] delayed heartbeat timer finding in 4a0e4157..HEAD:

• keepAliveStartTimeout is now tracked and cleared in stopKeepAlive() alongside the interval.
• Added fake-timer coverage: shutdown before 50ms priming delay emits no machine-alive; shutdown after first emit stops further ticks.

Waiting on CI re-run.

[github-actions]

Findings

• No findings.

Summary

• Review mode: follow-up after new commits
• No blocking issues found in the latest diff. The prior heartbeat timer leak appears addressed by tracking and clearing keepAliveStartTimeout.
• Residual risk: validation here was static because bun is unavailable in this runner.

Testing

• Not run (automation; bun not found in this environment).

HAPI Bot

[github-actions]

Findings

• [Minor] Bound machine metadata width in the sidebar — the new OS/host metadata is rendered as a shrink-0 flex item, so a long hostname or raw platform string can consume more than the sidebar width and push the health chip/session count off-screen. This is introduced in the new machine header row at web/src/components/MachineGroupHeader.tsx:100.
  Suggested fix:

  <span className="min-w-0 flex-1 truncate text-sm font-semibold text-[var(--app-fg)]">
      {props.label}
  </span>
  <span className="min-w-0 max-w-[8rem] shrink truncate text-[11px] text-[var(--app-hint)]">
      {machineMeta}
  </span>

Summary

• Review mode: follow-up after new commits
• One minor UI regression risk found in the latest diff. Residual risk: validation here was static because bun is unavailable in this runner.

Testing

• Not run (automation; bun not found in this environment).

HAPI Bot

[github-actions]

Findings

• [Minor] Make the health tooltip reachable on keyboard focus — the health chip is rendered inside the machine header <button>, but MachineHealthIndicator only relies on group-hover via HoverTooltip; unlike the existing session-row tooltip usage, it does not pass a parent-focus reveal class. Keyboard users can focus the row but cannot reveal the capacity/status/load tooltip introduced here. Evidence: web/src/components/MachineHealthIndicator.tsx:158.
  Suggested fix:

  // MachineHealthIndicator.tsx
  export function MachineHealthIndicator(props: {
      presentation: MachineHealthPresentation
      className?: string
      layout?: 'stack' | 'inline'
      compact?: boolean
      revealOnParentFocusClass?: string
  }) {
      // ...
      return (
          <HoverTooltip
              id={tooltipId}
              target={chip}
              side="bottom"
              align="row"
              className="shrink-0"
              tooltipClassName="px-3 py-2"
              revealOnParentFocusClass={props.revealOnParentFocusClass}
          >
              <MachineHealthTooltipBody presentation={presentation} />
          </HoverTooltip>
      )
  }
  
  // MachineGroupHeader.tsx
  <button className={cn('group/machine-row relative flex w-full min-w-0 ...')}>
      <MachineHealthIndicator
          presentation={props.healthPresentation!}
          layout="inline"
          compact
          className="shrink-0"
          revealOnParentFocusClass="group-focus-visible/machine-row:opacity-100 group-focus-visible/machine-row:visible"
      />
  </button>

Questions

• None.

Summary

• Review mode: follow-up after new commits
• One minor accessibility regression found in the latest diff. The previous metadata-width finding appears addressed by the follow-up commit. Residual risk: static review only; local automation could not run because bun is not installed in this runner.

Testing

• Not run (automation; bun not found in this environment).

HAPI Bot

[github-actions]

Findings

• No issues found.

Questions

• None.

Summary

• Review mode: follow-up after new commits
• No blocker/major/minor findings in the full current diff. The previous keyboard-focus tooltip finding appears addressed in the follow-up commit. Residual risk: static review only; I did not rerun the full Bun test suite in this review pass.

Testing

• Not run (automation)

HAPI Bot

[github-actions]

Findings

• No issues found.

Questions

• None.

Summary

• Review mode: follow-up after new commits
• No blocker/major/minor findings in the full current diff. I rechecked the CLI heartbeat health sampling, hub validation/broadcast path, SSE machine cache updates, and sidebar presentation wiring. Residual risk: static review only; I did not rerun the full Bun test suite in this review pass.

Testing

• Not run (automation)

HAPI Bot