Files
nixos-config/home/yolo/services.nix
T
Miguel Palhas 1ce7046994
CI / lint (push) Successful in 42s
CI / eval (push) Successful in 2m38s
fix(aoe-web): pin TMUX_TMPDIR and KillMode=process
Shells set TMUX_TMPDIR to the runtime dir; the unit did not, so the
daemon ran a second tmux server under /tmp. Every boot resumed duplicate
agents there, the dashboard diverged from the TUI, and polling the
invisible sessions fired an error push per session. Same fix pr-daemon
got.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-21 20:32:36 +01:00

110 lines
3.9 KiB
Nix

{
pkgs,
inputs,
...
}:
# The user services this box exists to run, ported from hand-written units in
# ~/.config/systemd/user on the Ubuntu machine.
#
# NOT self-contained: every ExecStart under ~/.bun or ~/.local/bin is an
# imperatively-installed binary, and the WorkingDirectories are clones of
# separate repos. Nix owns the unit definitions here, nothing more.
let
# A user unit gets almost no PATH by default; these are the profile dirs the
# original units got for free from the system PATH on Ubuntu.
toolPath = "%h/.local/bin:%h/.nix-profile/bin:/etc/profiles/per-user/naps62/bin:/run/current-system/sw/bin";
sem = inputs.sem.packages.${pkgs.system}.default;
aoe = inputs.agent-of-empires.packages.${pkgs.system}.aoe-with-web;
in
{
home.packages = [
sem
pkgs.bun
];
systemd.user.services = {
rev = {
Unit = {
Description = "rev always-on local code review server";
After = [ "network.target" ];
# MUST stay 0: at RestartSec=2 a fast-crashing rev burns the default
# 5-starts-per-10s budget, and systemd parks the unit in `failed` until
# a manual `systemctl --user reset-failed`.
StartLimitIntervalSec = 0;
};
Service = {
Type = "simple";
WorkingDirectory = "%h/tea/rev";
# nodejs_26, not pkgs.nodejs: rev's package.json sets engines >=26 and
# the nixpkgs default is 24.
ExecStart = "${pkgs.nodejs_26}/bin/node server/index.ts";
Environment = [
"NODE_ENV=production"
"REV_ROOTS=%h"
"REV_DEPTH=3"
"REV_SEM_BIN=${sem}/bin/sem"
"PATH=${toolPath}"
];
Restart = "always";
RestartSec = 2;
};
Install.WantedBy = [ "default.target" ];
};
rev-deploy = {
Unit = {
Description = "rev-deploy Gitea webhook listener that deploys rev on push to main";
After = [ "network.target" ];
# Same restart-budget trap as `rev` above.
StartLimitIntervalSec = 0;
};
Service = {
Type = "simple";
WorkingDirectory = "%h/tea/rev";
ExecStart = "${pkgs.bun}/bin/bun scripts/deploy-webhook.ts";
EnvironmentFile = "%h/.config/rev/deploy.env";
# Unit files are home-manager symlinks; deploy.sh must not rewrite them.
Environment = [ "PATH=${toolPath}" "REV_SKIP_UNIT_INSTALL=1" ];
Restart = "always";
RestartSec = 2;
};
Install.WantedBy = [ "default.target" ];
};
aoe-web = {
Unit = {
Description = "aoe serve Agent of Empires web dashboard";
After = [ "network.target" ];
# Same restart-budget trap as `rev` above.
StartLimitIntervalSec = 0;
};
Service = {
Type = "simple";
WorkingDirectory = "%h";
# aoe refuses `--auth none` on a non-loopback bind unless --behind-proxy
# is set; --allowed-host is what makes the rebinding gate accept a
# hostname under a wildcard bind (an IP literal needs no flag).
ExecStart = "${aoe}/bin/aoe serve --host 0.0.0.0 --port 8080 --auth none --behind-proxy --allowed-host aoe.n62.casa";
# TMUX_TMPDIR keeps the daemon on the same tmux server the shell and
# TUI use, instead of a second one under /tmp (same bug pr-daemon had).
Environment = [
"PATH=${toolPath}"
"TMUX_TMPDIR=%t"
];
Restart = "always";
RestartSec = 2;
# If this unit boots before any shell, the shared tmux server lands in
# its cgroup; the default control-group kill would take every session
# down on restart.
KillMode = "process";
};
Install.WantedBy = [ "default.target" ];
};
};
}
# pr-daemon, hourlog and week-review are not here: their units ship with the
# scripts they run, in the agent-skills module. This host opts in with
# programs.agentSkills.*.enable in home/yolo/default.nix.