[rangoDJ]

Fixes #65

What changed

• connecthub-init (new entry point): sends the container's entire output — every s6 service, the web server, start-up — through s6-log into /var/log/connecthub/all (rotated, 5 × 4 MB, timestamped), then execs /init, which stays PID 1. s6-log's 1 action passes each line on first, so docker logs is unchanged. If the log folder can't be written it just runs /init.
• s6-overlay's own catch-all logger (S6_LOGGING=1) was tried first in a test container on the current image: it doesn't start on this base, hence the wrapper.
• Client output: the RDP/VNC/SSH client's output is written to the container output tagged [RDP]/[VNC]/[SSH] (it used to go only to an in-memory session buffer). nginx's error log goes to stderr.
• Dashboard: the three tabs become one Logs window (last 2,000 lines, follows new lines while scrolled to the bottom) with a Filter box. Colours mark the session's client, ConnectHub's own messages, warnings and errors (including nginx's lowercase levels). One endpoint, GET /api/logs.
• Removed the separate Selkies log pipe (svc-selkies-log, its producer link and run-script prelude) — redundant now. CLIPBOARD_DEBUG → SELKIES_DEBUG moves into connecthub-init.
• README and compose updated.

Testing

• pytest: 252 passed, 8 skipped. Client output is tagged and printed; RDP errors in it are still reported; the polling filter covers /api/logs.
• Local dev server with a mixed log (Selkies, [RDP], server, xsettingsd, nginx): one window, no tabs; filtering on rdp shows only matching lines; classes as expected.
• Container: being tested from a manual build of this branch in a throwaway container (results in a comment).

🤖 Generated with Claude Code

[rangoDJ]

Container test of sha-1c8c26a (this branch) in a throwaway container on the Docker host, no ports/volumes/GPU:

• docker logs unchanged: LSIO start-up banner, init, Xvfb, xsettingsd, uvicorn and Selkies all present.
• /var/log/connecthub/all/current holds the same lines from the very first start-up message, timestamped by s6-log; mode 644, so the web server can read it.
• All services started normally with connecthub-init as the entry point (/init still PID 1 after exec).
  Container removed afterwards.