| name | cs16-server |
| description | Operate the Frag Fridays CS 1.6 server - run live console commands (change map, cvars) via the cmdpipe remote console, swap mods, restart, check status, tail logs, ship client config, manage maps/rotations, tune bots, verify plugins. Use for any request about the game server, the VPS, valve.zip, GunGame, deathmatch, bots or maps. |
CS 1.6 server operations
Browser CS 1.6 (Xash3D-FWGS WASM + WebRTC) on a Vultr VPS. Players join at
http://149.28.172.74:27016 - constant across all mods. SSH: cs16 alias
(verify with ssh cs16 true; if it fails, see memory note vps-access).
The prime directive: this stack fails silently - verify, don't assume.
Full gotcha list: docs/troubleshooting.md. Session procedure:
docs/runbook.md. Player-facing settings: docs/game-guide.md.
Layout
- Repo
server/ mirrors box /opt/cs16 1:1. Never hand-edit the box - edit
the repo and deploy. pnpm run pull re-syncs repo from box.
- Mods:
gg (GunGame), dm (deathmatch, our frag_dm.sma), kz (jump
maps, our kz.sma - checkpoints/timer, no bots), vanilla (root compose,
profile), zp (abandoned). One mod at a time - all bind 27016.
/opt/cs16/cs/ = game files tree (SteamCMD install, source of truth,
never in repo). /opt/cs16/valve.zip = the ONE canonical client payload,
mounted by every mod's compose.
Commands (from the repo root)
| Task | Command |
|---|
| Status | pnpm run status |
| Logs | pnpm run logs <mod> |
| Sync files only | pnpm run deploy |
| Swap/restart mod ("swap to gg/dm/kz/aim/vanilla") | pnpm run swap <vanilla|gg|dm|kz|aim> - announces the switch in-game first (when the running mod has cmdpipe), then runs deploy.sh. DROPS all players - never mid-session; still check for a sibling session first (memory: concurrent-sessions). Bare pnpm run deploy <mod> skips the announce |
| Ship client config / rebuild valve.zip | pnpm run clientcfg |
| Live server console (gg/dm/kz/aim only) | pnpm run rc "<command>" |
| Start a map vote ("votemap" / "vote map") | pnpm run votemap - 4 random mapcycle picks (script-side shuffle; never hand-pick maps). Run directly - no status/log checks first |
| Next map ("next map") | pnpm run nextmap - one shot: detects mod + current map, announces a 5s amx_csay warning, changelevels to the next mapcycle entry (wraps). Run directly - no status/log checks first |
| Announce ("announce " / "tell the players") | pnpm run announce "message" - green centre-screen HUD message to all players (amx_csay). Run directly - no status/log checks first |
| Restart map ("restart map" / "restart the map") | pnpm run restartmap - announces a 5s amx_csay warning, then changelevels to the current map (full reload: entities, timer, scores; players stay connected). Run directly - no status/log checks first |
| Rebalance teams ("rebalance" / "even the teams") | pnpm run rebalance - evens the T/CT headcount now (teambalance.amxx, gg/dm only): bots move first, then lowest-frag humans; moved players respawn instantly on the new side, nobody dropped. Run directly - no status/log checks first |
Iron rules
- After any swap:
docker ps must show exactly ONE container on 27016.
deploy.sh enforces this - never announce a mod without it. Containers look
identical in-browser (the port-theft incident).
- valve.zip root = only
valve/ and cstrike/. It is rebuilt by
update-clientcfg.sh from cs/{valve,cstrike} using a keep-list (union
of mod mapcycle.txt files; HL campaign maps always excluded). Client
changes (userconfig.cfg, new maps) need pnpm run clientcfg + players
hard-refresh. clientcfg RESTARTS the running mod at the end so the new
zip is served - it drops connected players; don't run it mid-session.
- Three registration files, not one: AMXX plugins ->
configs/plugins.ini;
AMXX modules -> configs/modules.ini; Metamod plugins (YaPB) -> Metamod's
own plugins.ini. A plugin absent from its file is a silent no-op.
- No ReAPI. No binary modules that sig-scan the CS DLL (killed CSDM).
Script-only Ham/fakemeta plugins work; engine-interface Metamod plugins
with explicit Xash3D support (YaPB) work.
- Configs are image-baked. yapb.cfg, gungame.cfg, mapcycle.txt,
plugins - all COPY'd at build. Editing repo files does nothing until
pnpm run deploy <mod> rebuilds. Only server/maps/*.bsp are live-ish
(compose-mounted into cstrike/custom/maps), but adding a map still
needs the full pipeline below.
Recipes
Run commands on the LIVE server: pnpm run rc "changelevel de_dust2".
No rcon exists on this stack (build answers no A2S/rcon UDP, stdin closed) -
rc.sh writes a serial-numbered file to /opt/cs16/cmdpipe/ which the
cmdpipe.amxx plugin (baked into gg + dm + kz + aim images, NOT vanilla) polls every
second. rc.sh tails docker logs for output, but slow output (map loads) can
outrun its 5s window - re-check with pnpm run logs <mod>. Map changes via
changelevel don't drop players; a redeploy does.
Interrogate a mod's console offline (image not yet deployed) - boot a
throwaway with stdin piped:
ssh cs16 '( sleep 10; echo "amxx plugins"; sleep 2; echo "quit" ) | \
docker run -i --rm --platform linux/386 <mod>-xash3d:latest +map de_dust2 +maxplayers 4 \
2>&1 | grep -aiE "bad load|running|plugins,"'
Also useful: meta list, amxx modules, status, changelevel <map>.
Use grep -a - some files/output trip binary detection.
Add a map: (1) verify the download is GoldSrc - first 4 bytes
1E 00 00 00; mirrors serve Source VBSP files under CS 1.6 names.
(2) Check dependencies (embedded textures vs wad refs vs skyname - python
lump-reader in the decision log; all-stock deps = free; check the box's
game tree before bundling anything, several "custom" skies/sounds are
stock). Non-map client assets (wads, overviews, skies, sounds) go in
server/custom/ - deploy.sh overlays it onto cs/cstrike/ and they ride
valve.zip; the server never needs them (verified: wad-referencing maps
boot without their wads). A missing skyname can be byte-patched to a stock
sky (recipe in the decision log). (3) Drop .bsp in
server/maps/, add to server/<mod>/mapcycle.txt. (4) pnpm run deploy <mod> then pnpm run clientcfg. (5) Boot-test in a throwaway with the
custom-maps mount and confirm: graph loads, bots rack up kills, zero
A* Search failed (a broken graph dead-airs the session - the he_glass
lesson). Boot map = compose command: +map - keep it equal to
mapcycle line 1.
Bots: YaPB config at server/<mod>/addons/yapb/conf/yapb.cfg (per-mod
on purpose - dm has yb_csdm_mode 1). Verify after restart: logs show
YaPB ... @ Xash3D Engine + Connecting Bot... lines. Unknown maps are
fine - YaPB downloads community graphs at map start.
Client config (binds, rates, crosshair): edit
server/config/userconfig.cfg, pnpm run clientcfg. The join binds
(F1/F2) live here - they are the only way players join (team menu doesn't
render in-browser).
Session day
A cron watchdog on the box (/opt/cs16/sim-watchdog.sh, source
server/sim-watchdog.sh) restarts the game container by itself when the sim
has killed itself (MAX_MODELS precache leak) or has been up >8h with nobody
connected - so an unexplained young container is probably it, not a sibling
session. Check /opt/cs16/logs/sim-watchdog.log.
Follow docs/runbook.md: bring up the mod, docker ps check, three Slack
messages (morning / midday "open the URL now to preload ~300MB" / final call
with F1-F2 + /guns instructions) - generate the text with the
/friday-posts skill. Don't swap mods mid-session - forces
every player through a reload. Mid-session map changes are fine:
pnpm run rc "changelevel <map>" keeps everyone connected.
