v0.58.0

Added

• Generalised the tmux wake transport to any terminal coding agent — Codex,
  Kimi K2, Claude Code — through synapse agent-tmux {start,wake,status,wait}
  with --agent-command. The pane-activity probe now derives the agent binary
  from the launch command instead of hard-coding Codex, so a non-Codex agent
  running under a shell is detected correctly. synapse codex-tmux remains as a
  Codex-defaulted alias (--codex-command); codex_tmux stays importable as a
  compatibility surface over the new agent_tmux module.

Fixed

• The connection-failure classifier now disambiguates the close codes the hub
  reuses. Code 4010 is emitted for both a takeover (superseded) and an
  authentication refusal (auth denied/auth required), and 4014 for both a
  takeover cooldown and the unauthenticated-socket cap; the classifier keyed on
  the code alone, so a bad token was reported as a takeover. It now reads the
  reason text, and recognises the auth-timeout (4012) and per-host-cap (4015)
  closes as well.
• The agent wake loop's retry backoff now adds bounded random jitter so a fleet
  of wakers that all lose the hub at once — a hub restart — does not reconnect in
  one synchronised burst.

Repository hygiene

• The PyPI publish workflow now triggers on the release tag push instead of
  release: published. A GitHub Release created by GITHUB_TOKEN does not fire
  downstream workflows, so the publish never ran automatically and each release
  had to be pushed to PyPI by hand; the tag push fires it directly.