[nsleigh]

Problem

After the computer sleeps or the network drops, the terminal often stops at
"Press ⏎ to Reconnect", and pressing Enter fails again. Only leaving the
dashboard and coming back works.

ttyd 1.7.7 retries a dropped connection once, straight away, then gives up.
That one try usually comes too early:

• After sleep, Home Assistant's ingress session has often expired. Every
  request from the panel gets a 401 (the Supervisor log shows "No valid
  ingress session") until Home Assistant's page creates a new session.
• During a Home Assistant restart, the add-on can't be reached at all (502).
• A clean close (code 1000) is never retried by ttyd.

Fix

New image-service/public/terminal-reconnect.js, installed by index.html:

• It watches the terminal frame for ttyd's "Press ⏎ to Reconnect" overlay.
• While that shows, it checks the add-on's /health after 1, 2, 5, then
  every 10 s. It also checks at once when the browser comes back online or
  the tab is shown again.
• When /health answers, it reloads the terminal frame, which reattaches to
  the same tmux session, so nothing running in it is lost.
• Each step is logged to the browser console as [terminal-reconnect].

Also removes --client-option reconnect=5 from run.sh. ttyd 1.7.7 has
no reconnect option (only disableReconnect), so it did nothing.

Home Assistant restarts run by Claude

This helps most when Claude restarts Home Assistant itself, for example
after editing the configuration it checks it and runs ha core restart.
Before, the terminal dropped with "Press ⏎ to Reconnect" and you had to press
Enter (or reopen the panel) once Home Assistant was back, then see what
Claude had done meanwhile. Now the terminal comes back by itself as soon as
Home Assistant is up, showing Claude's session where it left off, so you can
watch it carry on and check the result. The same applies to a restart
started from Settings while the terminal is open.

Tests

tests/test-terminal-reconnect.js (7 tests, added to run-tests.sh) covers:
spotting the overlay, the retry delays, checking at once on online or a
shown tab, no reload if Enter already reconnected, and watching the new
document after a reload.

Tested

On Home Assistant OS, through ingress:

• Home Assistant restart: retried through 502s while Core was down,
  then reloaded by itself. The released version needed Enter.
• Sleep with an expired session: /health got 401 until Home
  Assistant's page created a new session, then the terminal reloaded by
  itself. This took between about 8 and 30 s: Home Assistant's page checks
  its session once a minute (ha-panel-app.ts), so the wait depends on
  when that check falls after waking.
• A short sleep or Wi-Fi drop: ttyd's own retry still handles these; the
  new code doesn't act.

🤖 Generated with Claude Code

https://claude.ai/code/session_018rxd5pKHfU3wFULG6yLDE6

[ESJavadex]

Thanks @nsleigh for this fix and for testing the sleep, network-drop and Home Assistant restart scenarios! The automatic recovery is a useful improvement.

I added commit 43bb363 to cover two additional edge cases: health requests now time out and abort after 10 seconds, and recovery keeps retrying if the iframe navigation fails after a successful health check. Three new regression tests cover stalled requests, error pages and navigation that never finishes.

Validation passed: all regression suites in a Linux container (including the image-service integration tests), metadata checks, ShellCheck, JavaScript syntax checks, and Chromium/WebKit browser tests. The reconnect suite now has 10 passing tests.

The merge is currently blocked by the required GitHub Actions checks: GitHub rejects workflow dispatch with “Actions has been disabled for this repository”, despite the Actions settings reporting enabled. The new commit has no CI checks yet. Auto-merge is also disabled. Once Actions can run again, the required checks can validate the commit and unblock merging.

[nsleigh]

Thanks for adding this. Both are real gaps: a hung /health request would
have stopped the retries, and nothing retried after a reload that didn't
bring the terminal back.

I've pulled 43bb363 and the reconnect suite (10 tests) and the full
regression suite pass here too. I'll rerun the sleep and HA restart tests on
my install with this commit and report back.

One small thing, not blocking: if /health answers but the terminal page
keeps failing to load (e.g. ttyd itself has died), the watchdog now reloads
the frame every 10 s for as long as the panel is open. Before, it left the
error page alone. It's cheap, but using the same 1/2/5/10 s steps (or
stopping after a few reloads) would avoid a steady reload loop. Happy to
leave it as is if you prefer.