Repository navigation
Survey: laptop services & proxies — declarative coverage and conversion backlog #2775
Description
Survey taken 2026-10-30 on the laptop (amd), ordered by Drew alongside the muse-proxy direct-mode NixOS pass. Cross-refs: #2753 (direct mode), #2104 (proxy-supervisor retirement / per-proxy HM units), #1752 (PM hand-unit gaps — see status note below), #883 (stale-tree ExecStart trap).
Phase A result (today's stack)
The muse-proxy / muse-direct stack was already declarative — no functional change was needed:
systemd.user.services.muse-proxyinhome/common/default.nix(Proxy service-model redesign: automatic flake services, retire proxy-supervisor ownership, restore consumer Wants #2104) generates exactly the live unit: evaluating this flake at origin/main yields ExecStart/nix/store/7bxcdg21iqjmiinxh0bl92kyjwrbmkdn-muse-proxy-start-unit, byte-identical to the running unit, andnixosConfigurations.amdevaluates clean end to end.- The unit runs
bin/muse-proxy-start.shfrom the live checkout (egress unpinned,META_API_KEYunset by the script); direct mode reads~/.config/muse/auth.jsonat runtime — the credential is never in the config. Themuse-directprovider lives in.opencode/opencode.jsoncwithbaseURL http://127.0.0.1:{env:MUSE_PROXY_PORT:-18747}/v1, the port defined once asmuseProxyPortinhome/common/default.nix. - PR docs(nix): record muse-proxy direct mode in its Home Manager unit (#2753) #2774 (merged,
4189b191) only extended the unit comment so the NixOS source of truth records direct mode; re-evaluation after the edit produces the identical ExecStart, so nothing is superseded at activation and no restart is implied. There is no remaining activation step for this stack beyond the next ordinary rebuild.
Headline finding: the retired fleet-proxy-supervisor's units are still alive
#2104 retired the supervisor and made the per-proxy Home Manager units the sole owners of the canonical ports. In reality, seven fps-* transient units (started via systemd-run, no supervisor process exists) are still running, and four of them hold canonical ports, so their declarative replacements crash-loop:
| Port | Listener today | Declarative unit state |
|---|---|---|
| 18742 clef | fps-clef-proxy-18742 (since 2026-10-28 20:57) |
clef-proxy auto-restart, 1,170 restarts, exit 1 |
| 18743 agy | fps-agy-proxy-18743 (since 2026-10-28 13:02) |
agy-proxy auto-restart, 1,168 restarts, exit 1 |
| 18745 antigravity-ls | fps-antigravity-ls-proxy-18745 (since 2026-10-28 20:57) |
antigravity-ls-proxy auto-restart, 1,169 restarts, exit 1 |
| 18749 claude-code | fps-claude-code-proxy-18749 (since 2026-10-28 13:05) |
claude-code-proxy auto-restart, 1,169 restarts, exit 1 |
| 19744 / 19746 / 19748 | fps-codex-proxy-19744, fps-design-code-proxy-19746, fps-commandcode-proxy-19748 (all since 2026-10-29 23:08) squat the +1000 alternates |
canonical ports correctly held by the declarative units |
The proxies themselves also still write ~/.local/state/<proxy>/endpoint.json (fresh timestamps, incl. muse-proxy at its 09:30 restart today) — the files are live state written by the proxy processes, not just stale supervisor records as the #2104 comment assumes.
Services & proxies inventory
Fleet proxies (all user units; code in bin/, run from the live checkout)
| Service | Port | Managed today | Recommendation |
|---|---|---|---|
| muse-proxy | 18747 | HM unit, home/common/default.nix; running, direct mode live |
Already declarative — done (#2774) |
| codex-proxy | 18744 | HM unit + hand drop-in codex-proxy.service.d/warm-override.conf overriding ExecStart to the checkout |
Remove the drop-in: post-#2104 the unit already runs the checkout, so the #480-era override is redundant |
| design-code-proxy | 18746 | HM unit, running (fps twin on 19746) | Already declarative; kill the fps twin |
| commandcode-proxy | 18748 | HM unit + hand drop-in commandcode-proxy.service.d/fix-execstart.conf pointing ExecStart at ~/git/code/bin/... — the other checkout |
Remove the drop-in: it is the #883 stale-tree trap in live form; the HM unit already runs ~/code/bin/commandcode-proxy-start.sh |
| opencode-route-proxy | 18750 | HM unit, running | Already declarative |
| clef-proxy | 18742 | HM unit declared, crash-looping behind fps transient | Stop the fps unit at a natural break; unit then owns the port |
| agy-proxy | 18743 | same | same |
| antigravity-ls-proxy | 18745 | same | same |
| claude-code-proxy | 18749 | same | same |
| proxy-supervise.mjs (fps) | — | Retired in flake (#2104); 7 transient units still running; config fossils in ~/.config/proxy-supervisor/*.bak/.unmanaged-20261026; bin/proxy-supervise.mjs.new tracked leftover |
Retire fully: stop transients, delete fossils after |
Core AI services
| Service | Port | Managed today | Recommendation |
|---|---|---|---|
| openchamber | 3000 (tailnet) | HM unit, home/common/default.nix + hand drop-in openchamber.service.d/external-opencode.conf (attaches to the external OpenCode server; sets OPENCODE_HOST/OPENCODE_SKIP_START and a password env var) |
Convert the drop-in into the flake (or fold into the unit) so the external-opencode arrangement survives a config regeneration |
| opencode serve (standing server) | 13420 | Bare process (session scope, PPID 1, started 2026-10-30 09:18 from a shell) | Leave alone operationally (session-bearing); backlog: give it a declarative unit at a natural break — it is the process OpenChamber and the fleet's sessions depend on |
| mcpx | 34543 | HM unit, home/common/default.nix (+ mcpx-reload path unit) |
Already declarative |
| t3code | 3773 | HM unit, home/common/default.nix |
Already declarative |
| openclaw-gateway | 18789 | HM unit, home/openclaw/default.nix |
Already declarative |
| desktop-commander-remote | — | HM unit, home/common/default.nix |
Already declarative |
| sync-models | timer | HM unit + timer, home/common/default.nix |
Already declarative |
PM stack & apps
| Service | Port | Managed today | Recommendation |
|---|---|---|---|
| pm-projector | — | User unit declared via nixos/services/pm-stack/default.nix |
Already declarative |
| pm-aggregator | 4870 (tailnet) | same | Already declarative |
| pm-dispatch | 4879 (tailnet) | same | Already declarative |
| pm-tunnel (cloudflared) | 20241 metrics | same — this is the machine's running cloudflared; nixos/services/cloudflared/ only documents manual tunnel setup |
Already declarative; tidy the cloudflared module comment to point at pm-tunnel |
| pm-git-sync, pm-dashboard-backup, pm-projector-reconcile, quota-gatherer | timers | Declared via pm-stack / laptop-backups modules | Already declarative |
| docker-pm-grist / docker-pm-kaneo / docker-pm-pad / docker-pm-postgres | 3000 / 7777 / 8484 area (tailnet; owning pids not visible without sudo) | System units, NixOS-generated from nixos/services/pm-apps/default.nix (oci-containers) |
Already declarative — #1752 looks substantially landed; verify and close or re-scope it |
| ready-for-agent | 6056 (0.0.0.0) | User unit declared via nixos/services/pm-apps/default.nix, binary from the pm-stack deployment dir |
Already declarative (see #1752 note above) |
| pm-kaneo (user unit) | — | Declared (pm-apps/pm-stack), timer-driven companion to the docker app | Already declarative |
Automation, backups, quota exporters (user timers/services)
| Item | Managed today | Recommendation |
|---|---|---|
| laptop-autocommit-{hatch,hatch-personal,openclaw}, laptop-repo-keeper, laptop-untracked-review | HM via nixos/home/laptop-autocommit.nix / laptop-repo-keeper.nix |
Already declarative |
| atlas-backup, openclaw-backup, supabase-backup, repo-archive-backup, cold-store-{nightly,drill}, agent-backup | HM via nixos/home/laptop-backups.nix / home/common/default.nix |
Already declarative |
| meta-quota-export (svc+timer) | Hand-made files in ~/.config/systemd/user/ |
Convert to HM (same pattern as sync-models) |
| cloudflare-quota-export (svc+timer) | Hand-made files | Convert to HM |
| dashboard-backup (svc+timer) | Hand-made files | Convert to HM (alongside pm-dashboard-backup) |
| oc-password-sync (svc+path) | Hand-made files | Convert to HM, or fold into the openchamber declaration |
| git-maintenance@ (svc+3 timers) | Hand-placed copies of git's upstream template | Leave alone (upstream template) or declare via HM for completeness |
| rustdesk (user svc) | Hand-made file (ExecStart from /etc/profiles/per-user/...) |
Convert to HM or leave; low value either way |
System services (NixOS)
sshd, tailscaled, docker, dnsmasq, id-primary (nixos/services/id-mail), power-guard (nixos/services/power-guard), display-manager/desktop stack — all NixOS-generated from this flake. Already declarative; no action.
Bare processes (no unit)
| Process | Port | Recommendation |
|---|---|---|
python3 -m http.server 8899 |
8899 on 0.0.0.0 | Identify owner; retire (world-bindable directory server) |
python3 /tmp/mock.py |
18777 | Retire (test mock) |
pkgs/id debug build demo (~/git/code/pkgs/id/target/debug/id --data-dir ~/id-demo-data) |
3210 | Retire or replace with the packaged build; demo data dir |
node ~/git/fleet/2900-394-proxy-c… |
36539 | Identify; retire or promote to a declared proxy |
/tmp/mcpxdbg/mcpx daemon --detached |
38481 | Retire (debug instance; the real mcpx is a unit) |
| opencode serve | 13420 | See Core AI services — leave alone, backlog a unit |
Fossils & leftovers (inert, delete after the fps stop)
~/.config/systemd/user/proxy-supervisor.{service,timer}.unmanaged-20261026,~/.config/proxy-supervisor/config.json.unmanaged-20261026+ twoconfig.json.bak-**.backup/*.bak-*unit files next to pm-aggregator, pm-git-sync, pm-projector, quota-gatherer, repo-archive-backup, ready-for-agent units (inert, but they confuse every future survey)- Tracked in git under
bin/:codex-proxy.mjs.pre-warm-20261024,codex-proxy.mjs.resume-build.bak,commandcode-proxy-start.sh.bak-20261025003957-netfix,muse-proxy.mjs.bak-20261024-flag-tolerance,muse-proxy.mjs.bak-20261024121323-laptop-deploy,proxy-supervise.mjs.new - Hand drop-in
app-com.mitchellh.ghostty.service.d/overrides.conf(desktop app; noted for completeness)
Conversion checklist
- Stop the 7
fps-*transient units at a natural break so the 4 crash-looping declarative proxy units take their canonical ports; confirmsystemctl --usershows exactly one owner per proxy port afterwards - Remove
codex-proxy.service.d/warm-override.conf(redundant post-Proxy service-model redesign: automatic flake services, retire proxy-supervisor ownership, restore consumer Wants #2104) - Remove
commandcode-proxy.service.d/fix-execstart.conf(Home-layer eval uses a stale bin/ tree: commandcode-proxy unit 203/EXEC, managed opencode stuck on old rev #883 trap; points at the wrong checkout) - Fold
openchamber.service.d/external-opencode.confinto the flake declaration - Declare meta-quota-export, cloudflare-quota-export, dashboard-backup, oc-password-sync in Home Manager
- Verify nixos: bring Grist, Kaneo, Pad, ready-for-agent in as flake services — the last hand-run PM servers #1752 against pm-apps (docker-pm-* + ready-for-agent are declared and running) and close or re-scope it
- Retire the stray bare processes (http.server 8899, /tmp/mock.py, mcpxdbg, id demo, fleet proxy experiment)
- Backlog: a declarative unit for the standing
opencode serve(:13420), landed only at a natural break - Delete the fossil unit/config files and the tracked
bin/*.bakleftovers - Fix the Proxy service-model redesign: automatic flake services, retire proxy-supervisor ownership, restore consumer Wants #2104 comment in
home/common/default.nix: endpoint.json files are still written by the proxy processes themselves
Sandbox (out of scope for conversion)
The sandbox VM is not NixOS-managed: its services are ~/tools/init.sh run_steps (mcpx-service, opencode-service, muse-proxy-service, agy/commandcode/claude-code proxy services, proxy-supervisor-service, …), several deliberately disabled or stubbed by Drew's 2026-10-25 pass. Listed only so this survey is complete; no conversion is proposed.
Method: systemctl --user/system unit + timer listings, ss -tlnp port→pid→cgroup mapping, unit FragmentPath/file-type checks, and flake-source grep on the laptop, 2026-10-30. Nothing was stopped, restarted, or reconfigured during the survey.
Acceptance criteria
- Each
fps-*transient unit squatting a canonical port (18742, 18743, 18745, 18749) is stopped and its declarative Home Manager replacement is running and stable (no crash-loop) on that port, or the item is split into per-proxy follow-up issues naming the blocker. - The +1000 alternate-port squatters (19744, 19746, 19748) are cleared or explicitly documented as intentional.
- Hand drop-ins and hand-managed units in the inventory (e.g. codex-proxy warm-override) are either converted to declarative config or recorded as deliberate exceptions with a reason.
- The endpoint.json state files are documented as live proxy-written state (the Proxy service-model redesign: automatic flake services, retire proxy-supervisor ownership, restore consumer Wants #2104 assumption that they are stale supervisor records is corrected wherever it appears).
- A post-change survey re-run shows no
fps-*transient proxy units remain, and the result is posted on this issue.
Metadata
Metadata
Assignees
Labels
Large: multi-day or open-ended; expect staged check-insInfrastructure / host-portability workIntake review done: acceptance criteria, links, and normalized labels addedImportant and time-sensitiveQueue worker lifecycle stateFiled directly by a humanQueue worker lifecycle stateQueue packet kind: bounded fix with reproduction and targeted testsQueue worker lifecycle state
Activity
dezren39 commentedon Oct 30, 2026
Checklist item done 2026-10-30 ~10:06 CDT (Drew approved in chat): all 7 fps transient units stopped — fps-clef-proxy-18742, fps-agy-proxy-18743, fps-antigravity-ls-proxy-18745, fps-claude-code-proxy-18749, fps-codex-proxy-19744, fps-commandcode-proxy-19748, fps-design-code-proxy-19746. Verified after the stop: zero fps-* units remain; the Home Manager units clef-proxy / agy-proxy / antigravity-ls-proxy / claude-code-proxy are active/running on their canonical ports and their NRestarts counters are frozen (1319/1317/1318/1318 — cumulative incl. the crash-loop era, no longer climbing); ports 18742/18743/18745/18746/18747/18749 all listening; /health OK on 18742, 18747, 18749. No supervisor timer/service exists to resurrect the transients. muse-proxy (18747) untouched throughout.
dezren39 commentedon Oct 30, 2026
Drop-in items cleared 2026-10-30 ~10:14 CDT (Drew approved in chat). All three hand drop-ins moved out of systemd's view (recoverable copies in ~/.config/systemd/user/_dropin-backup-2026-10-30/) + daemon-reload: (1) commandcode-proxy fix-execstart.conf — verified the HM wrapper execs $HOME/code/bin/commandcode-proxy-start.sh (correct checkout); unit restarted onto it: active/running, NRestarts=0, :18748 /health OK. (2) codex-proxy warm-override.conf — HM wrapper already execs the same $HOME/code/bin/codex-proxy.mjs; running process was already equivalent, left undisturbed, still running. (3) openchamber external-opencode.conf — the fold-in already existed: bin/openchamber-start.sh (invoked by the HM unit) exports OPENCODE_HOST (default http://127.0.0.1:13420), OPENCODE_SKIP_START=true and OPENCODE_PASSWORD (file-sourced); live process env confirmed all three, so the drop-in was pure duplication. OpenChamber not restarted. Remaining checklist: the other hand-made unit groups + fossil cleanup.
dezren39 commentedon Oct 30, 2026
Intake review
🤖 Agent note · handed off
Still live? Yes. The body itself records PR #2774 as comment-only (no functional change, no activation step), and the headline finding — seven
fps-*transient units still holding canonical/alternate ports while their declarative replacements crash-loop (~1,170 restarts each) — is a survey result from today, not a fixed item. No merged PR resolves it.Notes / risks: The crash-looping declarative units are the urgent slice: four canonical ports are served by units #2104 believed retired. Risk is change-window risk on the laptop (stopping a transient unit before its replacement binds cleanly drops that proxy). Suggested approach: one proxy at a time — stop the
fps-*unit, start the HM unit, verify the port and endpoint.json — starting with one low-stakes proxy before repeating the pattern.Suggested approach: treat the acceptance criteria as a checklist; split per-proxy issues only if a single conversion blocks on something specific.
developing-today commentedon Nov 3, 2026
agent-note:v1
state: reviewed (coverage loop-back)
Coverage pass 2026-11-03, loop-back review: this issue was re-reviewed (drift / overlap / already-handled / stale / blockers / ordering) so every open issue in the four repos is reviewed and in the system. Verdict: still live; in the system as task 7358 (triage, queued, priority 50) in repo-health-code-pilot. Labels audited and synced. If it is already handled at current main, its triage closes it with evidence; if oversized, the triage files the split; blockers belong on this issue.
Found via: coverage loop-back 2026-11-03
developing-today commentedon Nov 3, 2026
agent-note:v1
state: reviewed (coverage loop-back)
Coverage pass 2026-11-03, loop-back review: this issue was re-reviewed (drift / overlap / already-handled / stale / blockers / ordering) so every open issue in the four repos is reviewed and in the system. Verdict: still live; in the system as task 7358 (triage, queued, priority 50) in repo-health-code-pilot. Labels audited and synced. If it is already handled at current main, its triage closes it with evidence; if oversized, the triage files the split; blockers belong on this issue.
Found via: coverage loop-back 2026-11-03
dezren39 commentedon Nov 3, 2026
🤖 Agent note · coverage finding routed here from the #2104 triage (queue task 7290, Lane P, 2026-11-03)
One stale-reference item for this survey's declarative-coverage backlog, found while verifying #2104 (the proxy service-model redesign, PR #2107, merged 2026-10-30) live:
home/openclaw/default.nix(current main): the comment abovesystemd.user.services.openclaw-gatewaystill says the provider proxies "are owned by the proxy-supervisor (transient fps-* units on the canonical ports)" and that the gateway "must not Want the manual per-proxy units" — both false since feat(home): make fleet proxies automatic services, retire proxy-supervisor (#2104) #2107 retired the supervisor and made the proxy units real automatic services.openclaw-gateway'sAfterlist still namesagy-proxy.service, which no longer exists as a unit (agy-proxy was replaced/retired in the later antigravity rework — seehome/common/default.nix:588-589). On the laptop it surfaces asagy-proxy.service not-found inactive deadinsystemctl --user list-units "*proxy*".After-only, so it is harmless ordering noise, but it is exactly the kind of stale declarative reference this survey tracks.No separate issue filed — this survey is the tracking home. #2104 itself was closed with evidence in the same pass.
42 remaining items
dezren39 commentedon Nov 5, 2026
agent-note:v1
event: picked_up
task: 8983
worker: repo-worker-code
lane: 02cafe76-eafd-4d0b-aaa6-2b7a737d68dc
state: claimed
verified: Fresh GitHub state: #2775 OPEN; no held/user-owned/do-not-merge/do-not-auto-edit labels; PR #2107 is MERGED; no open PR search result for #2775. Claim labels synced to task/small-fix, queue/claimed, status/in-progress, work/assigned.
next: Fast-forward the laptop primary main, create or resume the task worktree, then make only the scoped comment/dependency corrections.
at: 2026-11-05T07:02:46Z
dezren39 commentedon Nov 5, 2026
agent-note:v1
event: pr_opened
task: 8983
worker: repo-worker-code
lane: 02cafe76-eafd-4d0b-aaa6-2b7a737d68dc
state: claimed; PR #6113 OPEN
verified: Changes are on queue/8983-2775-proxy-comment-cleanup at 1c2d261; targeted Nix parses/evaluation and git diff --check pass. The obsolete agy-proxy After entry is removed while commandcode/codex ordering remains.
next: Inspect fresh PR checks; if pending beyond the five-minute active window, park task 8983 with CI resume condition.
dezren39 commentedon Nov 5, 2026
🤖 Agent note · handed off
dezren39 commentedon Nov 5, 2026
🤖 Agent note · handed off
ses_e5e5503c176ffegaHZ1Urk2VOVdezren39 commentedon Nov 5, 2026
agent-note:v1
event: task_complete
task: 8983
worker: repo-worker-code
session_id: ses_e5e5503c176ffegaHZ1Urk2VOV
state: scoped correction ready in PR #6113; issue #2775 remains OPEN for the broader inventory/conversion checklist
result: Corrected Home Manager ownership wording, documented live proxy-process publication/removal of endpoint.json, and removed the obsolete agy-proxy.service After edge. No runtime service behavior changed and no services were restarted.
branch: queue/8983-2775-proxy-comment-cleanup
head: e532ffa (includes main through e3626d1)
validation: git diff --check; nix-instantiate --parse home/openclaw/default.nix; nix-instantiate --parse home/common/default.nix; nix eval --raw .#nixosConfigurations.amd.config.home-manager.users.user.home.activationPackage.drvPath — all pass
CI: No GitHub Actions run exists for this head; CodeRabbit is rate-limited/neutral. The proxy-port-fallbacks workflow currently triggers only on merge_group, while nix-ci is schedule/workflow_dispatch only, so the policy-listed PR Nix-expression coverage is absent. Local targeted evaluation passed.