Implements the Linux half of AGENT_LOCAL_DISCOVERY_SPEC.md: when a local
theta-gateway/theta-proxy announces itself via mDNS as fronting this
agent's ServerURL host, skip the relay/WAN path and talk to it directly.
Off by default (config.PreferLocalDirectory / prefer_local_directory)
since it changes host name resolution.
- local_discovery.go: polls for _theta-suite._tcp every 30s via
hashicorp/mdns, matches the TXT "hosts" field against the agent's own
target host, applies/clears a hosts-file override on change. Presence/
absence of the mDNS announcement IS the "on this LAN or not" signal --
no separate network detection needed, since multicast doesn't cross
routers/VLANs.
- hosts_override.go: writes a single marked, idempotent block into
/etc/hosts (never touches anything else in the file); clearing removes
the block entirely rather than leaving empty markers.
- HARD RULE preserved: this only ever changes DNS resolution, never TLS
trust -- nothing here touches certificate validation, so a spoofed
rogue mDNS announcement produces a TLS failure against the real
hostname's cert, not a silent MITM.
Verified end-to-end with real containers (Node mDNS announcer + this
actual Go binary, not mocked), which caught two real bugs neither showed
up in code review:
1. mdns.Lookup()'s DefaultParams() requests both IPv4 and IPv6. The
underlying client sends the v4 query (which got a real, valid
response per a packet capture), then sends the v6 query, and if THAT
send fails (no IPv6 route -- common on plain v4 hosts/containers) the
whole Query() returns that error synchronously, before ever entering
the response-listening loop. The v4 response was silently discarded.
Fixed by building QueryParam manually with DisableIPv6: true instead
of using the Lookup() convenience wrapper.
2. The original hosts-file writer used write-tmp-then-rename for
atomicity. /etc/hosts is frequently a bind mount (every container
runtime does this) -- rename() onto a bind-mounted file fails with
EBUSY, since you can't atomically replace a mountpoint. Switched to
truncate-and-rewrite in place; the process-local mutex already
serializes writers, so the lost atomicity is a small, acceptable
tradeoff against a confirmed hard failure.
Full cycle verified: announcer starts -> agent discovers it -> hosts
override applied -> announcer stops -> override cleanly reverts, no
stale entry, no discovery trace left.
Windows/macOS remain unbuilt -- need platform-native testing this
environment can't do (see AGENT_LOCAL_DISCOVERY_SPEC.md §3's open
question: hosts-file edits vs. a local stub resolver, per-OS elevation
and DNS-cache behavior).
Second milestone of the Windows parity work (DESIGN-WINDOWS.md §13).
WireGuard mesh client (signed, WSS-delivered):
- wireguard_apply / wireguard_remove commands: Ed25519-verified, gated on a new
wireguard capability. Linux applies via wg-quick up/down; Windows installs the
peer config as a WireGuardTunnel service via wireguard.exe
(/installtunnelservice, /uninstalltunnelservice)
- state polling in the home monitor drives the blue tray icon and auto-VPN:
connect when away from home + auto_vpn, disconnect on return (2m cooldown)
- tray VPN toggle and the auto-VPN checkbox are now live; the preference
persists to agent.yml (PersistAutoVPN)
IAM on Windows (iam_windows.go):
- allowed_login_groups -> net localgroup; ssh_keys -> per-profile
authorized_keys + %ProgramData%\ssh\administrators_authorized_keys;
revoke_users -> helper logs off all of the user's WTS sessions;
sudo_rules logged as no direct equivalent
Tray enrichment:
- Open Config (opens agent.yml), Clear enrollment (re-enroll) menu items
- set_auto_vpn persists; vpn_connect/vpn_disconnect/reinit/open_config commands
handled by the daemon (tray_server.go)
Packaging & release:
- installer/windows/installer.iss: fully-offline Inno Setup bundle (agent,
tray, helper, vendor-signed WireGuard MSI, OpenCredential CP, VC++ redist),
/SILENT /SERVER_URL /JOIN_KEY parameters, SYSTEM service + HKLM Run tray
autostart, Users-writable %ProgramData%\Theta42 for the IPC socket
- .github/workflows/build-windows.yml: build + test, pinned vendor downloads,
ISCC compile, Azure Trusted Signing (OIDC), SHA256SUMS, GH release attach,
optional SSO resource publish
- agent.yml.example documents auto_vpn, wireguard, service_name,
desktop_helper, public_ip_detect
Tests:
- wireguard_apply/remove dispatch (allowed + capability-denied), PersistAutoVPN,
ClearEnrollment; dispatch tests pin linuxPlatformOps with a temp WireGuard conf
- end-to-end verified against a local mock SSO on Windows: join-key enrollment
(token persisted, join key blanked), discovery/telemetry pushed, signed
arbitrary_bash verified + executed via powershell -EncodedCommand; tray IPC
socket binds %ProgramData%\Theta42; LDAP byte-pump binds 127.0.0.1:389;
helper update swap verified
Rebuilds all tracked dist binaries (v2.1.0).
First Windows parity milestone (DESIGN-WINDOWS.md §13 build order item 1).
- Add a PlatformOps abstraction so command dispatch is OS-neutral:
- linuxPlatformOps keeps today's systemctl/journalctl/bash behavior (deliberately
untagged so shared dispatch tests run on Windows CI)
- windowsPlatformOps maps reboot/shutdown to shutdown.exe, service control to
sc.exe (stop+start for restart), fetch_logs to Get-WinEvent, arbitrary_bash to
powershell -EncodedCommand (byte-exact under arbitrary quoting), and declines
configure_ldap (Windows logon goes through OpenCredential)
- Run theta-agent as a Windows service (x/sys/windows/svc): SYSTEM auto-start,
SCM stop/shutdown handling; CLI install-service/remove-service via svc/mgr
- Add theta-agent-helper (session-0 companion): lock/display_off/logout via
user32/wtsapi32, and staged self-update (wait for service stop, swap the
locked exe, sc start)
- Self-update becomes platform-aware: Linux renames over the running binary;
Windows stages .new and hands the swap to the helper (running exe is locked)
- Platform paths: agent.yml and tray.sock under %ProgramData%\Theta42 (the
service runs as SYSTEM while the tray runs as the user, so the per-user temp
dir no longer works for tray IPC); LDAP byte-pump falls back to TCP loopback
- config: service_name, desktop_helper, public_ip_detect (air-gap: skips
external public-IP lookups in telemetry + home monitor), wireguard block
- cli: platform-aware config path + self-update artifact name + service restart
- tests: dispatch tests pin linuxPlatformOps; 0600 mode assertions gated to
POSIX so the suite is green on Windows
Rebuilds all tracked dist binaries (v2.1.0).
See CHANGELOG.md for the full breakdown. Summary:
- ldap_tunnel.go: serves a local unix socket for SSSD/PAM and relays raw
bytes to the SSO over the existing WSS channel (ldap_tunnel messages);
the agent never parses LDAP (DESIGN.md §4). Adds safeWriter to
serialize WebSocket writes now that telemetry, heartbeat, the LDAP
tunnel, and command responses all share one connection.
- secrets.go: renders local templates ({{ bao "path#key" }} placeholders)
by fetching node-scoped values from the SSO and writing the target
atomically at 0600, on a signed render_secrets command (DESIGN.md §5).
demo/ has minimal bash + Node consumers of the rendered file.
- iam.go: applies signed node IAM pushes -- sudoers.d rules (visudo -c
validated), SSH AuthorizedKeysCommand keys, /etc/security/access.conf,
and revocation via sss_cache -E + pkill -u (DESIGN.md §6).
- Capability reporting: the agent's enabled capabilities ride along in
its discovery frame so the SSO can show them in the Directory.
- DESIGN.md: the v2 protocol design this implements.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Installing the agent with one key is now all it takes to add a host.
New `join_key` config field, presented while auth_token is empty. The SSO
exchanges it for this agent's own token and the public key it must pin,
both delivered in the config frame; the agent persists them and blanks
the join key. Nothing has to be copied between two machines by hand.
PersistEnrollment rewrites only the credential lines -- line-based rather
than a YAML round-trip -- so operator comments, the capability matrix and
formatting survive. It re-reads afterwards so the credential is live
without a restart, and keeps the file 0600.
The connect URL carries ?hostname= so a self-enrolling host is named
after itself, and the agent refuses to connect at all (with a long
back-off) when it has no credential rather than presenting an empty one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
wmantly