0.8.38

Release Notes

From the first field reports on 0.8.37: the plugin now follows a resumed session, never hands a session another agent's hubs, and has confer plugin status|restart. Detached watchers upgrade themselves too.

• A resumed session arms for the right session. After a Claude Code resume the shell can keep
  the pre-resume CLAUDE_CODE_SESSION_ID while the plugin reader runs under the real one. Every
  confer arm stamped the stale id, the reader delivered nothing, and the agent fell back to
  30-minute Monitors all night (jarvis). When the shell's id has no live reader but the Claude Code
  process that runs it has one, confer now uses that reader's session, and arm says so.

• A fresh session no longer inherits another agent's hubs. The plugin remembered each
  project's hubs so a new session could resume without arming, but it remembered only the last
  agent to use the directory, so a new jarvis session was handed herdr's hub (jarvis). Once a
  session arms, remembered hubs of other roles are dropped. When two agents with no role in
  common have used a project, fresh sessions there wait for their own arm.

• A watcher that cannot start names the binary. A reader still running a binary an upgrade had
  replaced failed every spawn with ENOENT, and the error blamed the clone directory. The error now
  names both, and on Linux confer plugin status flags a reader whose binary was replaced on
  disk. Detached watchers are now started by the path confer was launched as, not the resolved
  executable.

• A detached spool watcher re-execs when its binary is replaced. The plugin reader already does
  this (0.8.37). On Grok the watcher is reparented to pid 1 and outlives the Monitor, so an upgrade
  left it on the old build until the next confer arm. It now execs the new confer in place: same
  pid, the lock is adopted instead of signalling itself, and it prints one upgraded from … line.
  Inline watchers are unchanged. CONFER_WATCH_UPGRADE_SECS (default 30) is the check interval.

• confer plugin status and confer plugin restart. The documented way to move an old plugin
  reader onto a new build was to read its pid from ~/.confer/plugin/readers/<session>.json and
  kill it. On a real machine that file named a reader that had died hours before, and the live
  reader for the project ran under a different session id, so the kill hit nothing
  (batcave-net). status lists the running readers and marks yours. restart stops yours (this
  session's, else this project's; --pid picks one when several serve a project), then waits for
  the plugin to start a fresh one on the installed confer and reports it. It only ever signals a
  pid that a reader record names and that is a live confer process.

• Reader records are cleaned up. A reader removes its record when it exits, and a starting
  reader clears records of readers that are no longer running.

• A deliberate stop restarts in 2 seconds, not 30. The plugin's wrapper script now tells a
  clean stop (exit 0) from a failure; failures still back off for 30 seconds. Takes effect when
  Claude Code updates the plugin.

• A watcher's startup lines are no longer delivered as wakes. "--replace killed the existing
  watcher", "owned by role", "streaming new items" and the like reached the agent as six
  notifications every time a watcher restarted (astrolabos-voice). They still go to the spool
  file for debugging.

Install confer-cli 0.8.38

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/codeshrew/confer/releases/download/v0.8.38/confer-cli-installer.sh | sh

Install prebuilt binaries via Homebrew

brew install codeshrew/tap/confer

Download confer-cli 0.8.38

File	Platform	Checksum
confer-cli-aarch64-apple-darwin.tar.xz	Apple Silicon macOS	checksum
confer-cli-x86_64-apple-darwin.tar.xz	Intel macOS	checksum
confer-cli-aarch64-unknown-linux-musl.tar.xz	ARM64 MUSL Linux	checksum
confer-cli-x86_64-unknown-linux-musl.tar.xz	x64 MUSL Linux	checksum