A member's tmux server keeps the version it started with until their last
pane closes, but the dashboard asked it with TMUX_BIN, the newest tmux: a
3.8 client on a 3.6 server spun it at 100% CPU (three members needed a
manual kill on 2026-10-15).
- /opt/agenthub/libexec/tmux-member (install-system, in sudoers): runs a
member's tmux call with the binary of THEIR running server
(/proc/<pid>/exe, found from the socket in /proc/net/unix), else
TMUX_BIN. It runs as the member and execs only a process they own.
sudoArgv uses it once installed (linux-accounts readConf), TMUX_BIN
until then.
- Users → Members' tmux (GET /api/users/tmux): each member server's version
beside the installed one, sessions, agents working. An admin restarts one
(POST /api/users/:id/tmux-restart) or every idle outdated one
(POST /api/users/tmux-restart-idle), with their password again, audited
per member; a server that will not stop answers 409 ETMUXSTUCK naming
its pid for root.
- server-s3-system.test.js: `delete` is a provision action since Delete
user, so the "unknown action" case now uses `purge`.
Operators: run `sudo agenthub server install-system` once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- self-restart: a server in our unit's cgroup (/agenthub.service, system
or user) restarts via `systemctl restart --no-block` (sudo -n for the
system unit), our launchd job by exiting into KeepAlive. The detached
restart-helper replacement held the port while Restart=always started
the unit every 3 s (v2.7.2 on the S3 box: 240+ restarts, MainPID 0).
- run-as.mayHaveTmux: only ENOENT means "no tmux". A dev server in server
mode without CAP_DAC_READ_SEARCH got EACCES on a member's 0700
.agenthub/tmux, dropped the member from the list and ended their runs
within seconds — the real cause of `401 not a live pane's token`.
- tmux: a member's list is re-asked every 5 s, not 1 s (one sudo + three
auth.log lines each).
- POST /api/lanes/:id/start: 409 ENOCWD for a folder that no longer
exists (the pane started in the account's home otherwise).
- Settings: Remote sources only for staff (a developer got a 403).
- scripts/test-server.js drops every inherited AGENTHUB_* (pane env).
Tests: self-restart.test.js, server-s3-panes (mayHaveTmux), lanes-api
(ENOCWD); touched suites 771 + client pages 191, tsc -b.
A frozen member (SIGSTOPped tmux server) cost every user one synchronous
3 s block of the event loop per minute (F37's back-off ceiling). Member
listing now runs through async execFile (same 3 s / SIGKILL / 1 MiB
bounds, same back-off) into the listed cache the synchronous readers
(liveSessions, listSessions, listOutput) already use: stale-while-
revalidate, 15 s max age, cold or failing = that member's panes UNKNOWN
(F38 unchanged). A session this process creates or kills is written into
the member's cached answer at once, and a list in flight across that
change is dropped. Host tmux calls, local mode and every click's call
(new-session, send-keys, kill, has-session) are unchanged.
review5's stalled-member test now polls for 4 s past the timeout and
asserts each poll < 100 ms, event-loop stall < 500 ms and one sudo
(old code: first poll 3157 ms).
Server mode can now run each account as its own Linux user (ah-<name>)
when `agenthub server install-system` has set S3 up. The dashboard
reaches a member only through sudo -u / write-as / pane-launch, reads a
member's files only through owner- and type-checked opens, and hooks
are authenticated by peer uid. Local mode and server mode without S3
are unchanged.
EXPERIMENTAL: tested with fake sudo and fake accounts only. It has not
yet been verified on a real install with real member accounts, sudo or
a real Codex login. Do not offer it to real users yet.
- S3 core: provisioning, per-member tmux servers, the proxy key
gateway, shared tools in /opt, dropped capabilities for children
(P0–P5)
- P4: plan and quota per member home
- P6: ~/.claude.json seed; opt-in plugin baseline installed as the
member; managed-settings spike script (not run)
- P7: Codex account pool (`agenthub accounts add --tool codex`)
- Roles merged with main's admin / analyst / developer: analysts write
only their own rows, under S3 as well
- Per-account prefs: one implementation (Chat UI + onboarding)
- Security reviews F1–F49 (five rounds), each fix with a regression
test: symlink/TOCTOU-safe member reads, no host writes in a member
home, credentials and the data dir never served, owner and folder
account must agree, pane tokens bound to the pane's uid, bounded
sudo/tmux/git calls and member reads (no hang or crash from a member),
async route rejections answer 500 instead of exiting
- Docs: docs/SERVER-MODE.md (S3 model, accepted ceilings, operator
runbook), docs/plans/server-mode/S3-PLAN.md, CLAUDE.md S3 rules
Tests: server 2838/2841 (3 need 127.0.0.1:3000, held on this box);
client 1162/1162; tsc clean; headers check passes.
Co-Authored-By: Claude <noreply@anthropic.com>
nntrivi2001