A lightweight system agent that collects server information and installed packages, then reports them to a VulTrack vulnerability management server.
VulTrack Agent was developed with AI assisted coding. Be aware there might be dragons.
The VulTrack Server can be found in this repository.
- Static Binary: No external runtime dependencies, easy deployment
- Flexible Configuration: Config file, environment variables, or command-line flags
- Automatic Package Detection: dpkg (Debian/Ubuntu) and rpm (RHEL/CentOS/Rocky/Alma)
- Robust Error Handling: Retry logic with exponential backoff for network and server errors
- Runtime Facts: Loaded kernel modules, listening sockets, packages in use, services and kernel settings let the server rule out vulnerabilities that cannot affect the host
- Automatic Token Refresh: Short-lived access tokens are refreshed automatically; the agent re-enrolls itself when the refresh token expires
- Systemd Integration: Pre-configured service for easy installation
- Ansible Role: Included Ansible role for automated fleet deployment
| Status | |
|---|---|
| Release artifacts | Linux amd64 only: static binary and .deb |
| Debian / Ubuntu | Supported (.deb, binary, Ansible role) |
| RHEL / CentOS / Rocky / Alma | Package collection works; install the static binary manually (there is no RPM package). Runtime facts that need dpkg are reported as unknown |
| arm64 | Not released; can be built from source (ARCH=arm64 ./build.sh binary) |
| Ansible role | Ubuntu (focal, jammy, noble), amd64 |
An Ansible role is included at ansible/roles/vultrack_agent/ for automated
installation, configuration, enrollment, and service management across a fleet
of servers.
For manual installation options, see INSTALL.md.
- Ansible 2.12+
- Target hosts: Ubuntu with
systemd
# Version to install
vultrack_agent_version: "1.0.0"
# Installation method: "deb" (recommended) or "binary"
vultrack_agent_install_method: "deb"
# VulTrack server URL — REQUIRED
vultrack_agent_server_url: ""
# Set to true to enroll the agent during the play
# When true, vultrack_agent_enrollment_key must also be set
vultrack_agent_enroll: false
# IMPORTANT: enrollment_key must remain in the config permanently.
# The agent needs it to re-enroll automatically when the refresh token
# expires or is revoked.
vultrack_agent_enrollment_key: ""
# Agent configuration
vultrack_agent_refresh_token_file: "/var/lib/vultrack-agent/refresh.token"
vultrack_agent_report_interval: "1h"
vultrack_agent_log_level: "info"
vultrack_agent_log_file: ""
vultrack_agent_insecure: false
vultrack_agent_ca_cert: ""
# Runtime facts (see "Runtime Facts" below)
vultrack_agent_collect_runtime: true
vultrack_agent_runtime_collectors: [kernel, listeners, packages_in_use, package_info, services, virtualization]
vultrack_agent_runtime_timeout_seconds: 20
vultrack_agent_module_index_state_file: "/var/lib/vultrack-agent/module-index.sha256"
# Systemd service management
vultrack_agent_service_state: "started"
vultrack_agent_service_enabled: true
# GitHub source (usually no need to change)
vultrack_agent_github_url: "https://github.com"
vultrack_agent_github_repo: "RiskIdent/VulTrack-Agent"
# Installation paths (usually no need to change)
vultrack_agent_bin_path: "/usr/local/bin/vultrack-agent"
vultrack_agent_config_dir: "/etc/vultrack-agent"
vultrack_agent_config_file: "/etc/vultrack-agent/config.yaml"- name: Deploy VulTrack Agent
hosts: all
become: true
roles:
- role: vultrack_agent
vars:
vultrack_agent_version: "1.2.1"
vultrack_agent_server_url: "https://vultrack.example.com"
vultrack_agent_enroll: true
vultrack_agent_enrollment_key: "{{ lookup('env', 'VULTRACK_ENROLLMENT_KEY') }}"- name: Deploy VulTrack Agent
hosts: all
become: true
roles:
- role: vultrack_agent
vars:
vultrack_agent_version: "1.2.1"
vultrack_agent_server_url: "https://vultrack.example.com"- name: Deploy VulTrack Agent (binary)
hosts: all
become: true
roles:
- role: vultrack_agent
vars:
vultrack_agent_version: "1.2.1"
vultrack_agent_install_method: "binary"
vultrack_agent_server_url: "https://vultrack.example.com"- Selects the release architecture (always
amd64; no other architecture is released) - Validates required variables (
vultrack_agent_server_url, andvultrack_agent_enrollment_keywhen enrolling) - Creates the system user and config directory (
/etc/vultrack-agent/) - Downloads and installs the agent from GitHub Releases (
.debor binary) - Writes the configuration file from template
- Optionally runs
vultrack-agent enroll(skipped if refresh token file already exists — idempotent) - Ensures the systemd service is enabled and running
Configuration is loaded from three sources in the following priority order (highest to lowest):
- Command-line flags
- Environment variables
- Config file
Default location: /etc/vultrack-agent/config.yaml
A custom path can be specified with the global --config flag:
vultrack-agent --config /path/to/config.yaml reportFull reference:
# VulTrack server URL (required)
server_url: https://vultrack.example.com
# Enrollment key (required — keep this permanently in the config).
# Used for initial enrollment and for automatic re-enrollment when the
# refresh token expires or is revoked.
enrollment_key: ""
# Path to persist the refresh token on disk.
# The access token is held in memory only and is never written to disk.
refresh_token_file: /var/lib/vultrack-agent/refresh.token
# How often to send reports in daemon mode (default: 1h)
# Examples: 30m, 1h, 2h, 24h
report_interval: 1h
# Log level and log file are validated but not yet used: the agent
# currently writes all messages to stderr (the journal under systemd).
log_level: info
log_file: ""
# Skip TLS certificate verification — only use for testing!
insecure: false
# Path to a custom CA certificate in PEM format (optional)
ca_cert: ""
# Runtime facts (see "Runtime Facts")
# Master switch; false omits the runtime facts entirely (default: true)
collect_runtime: true
# Collectors to run (default: all)
runtime_collectors: [kernel, listeners, packages_in_use, package_info, services, virtualization]
# Overall time budget for all collectors in seconds (default: 20)
runtime_timeout_seconds: 20
# Hash of the last kernel module index the server accepted
module_index_state_file: /var/lib/vultrack-agent/module-index.sha256An annotated example is available at contrib/config.yaml.example.
| Variable | Description |
|---|---|
VULTRACK_SERVER_URL |
VulTrack server URL |
VULTRACK_ENROLLMENT_KEY |
Enrollment key (keep permanently) |
VULTRACK_REFRESH_TOKEN_FILE |
Path to refresh token file |
VULTRACK_REPORT_INTERVAL |
Report interval (e.g. 1h, 30m) |
VULTRACK_LOG_LEVEL |
Log level (debug, info, warn, error); validated, not yet used |
VULTRACK_LOG_FILE |
Path to log file; not yet used |
VULTRACK_INSECURE |
Skip TLS verification (true/false) |
VULTRACK_CA_CERT |
Path to custom CA certificate |
VULTRACK_COLLECT_RUNTIME |
Collect runtime facts (true/false) |
VULTRACK_RUNTIME_COLLECTORS |
Comma-separated list of collectors (e.g. kernel,services) |
VULTRACK_RUNTIME_TIMEOUT_SECONDS |
Time budget for all collectors in seconds |
VULTRACK_MODULE_INDEX_STATE_FILE |
Path to the module index state file |
| Flag | Description |
|---|---|
--config |
Path to config file |
--server-url |
VulTrack server URL |
--refresh-token-file |
Path to refresh token file |
--log-level |
Log level; validated, not yet used |
--log-file |
Path to log file; not yet used |
--insecure |
Skip TLS verification |
--ca-cert |
Path to custom CA certificate |
These flags work with every command. Command-specific flags:
| Command | Flag | Description |
|---|---|---|
enroll |
--enrollment-key |
Enrollment key (otherwise enrollment_key from config/env) |
enroll |
--force |
Revoke an existing registration and enroll again |
export |
-o, --output |
Write the JSON to a file instead of stdout |
There are no flags for report_interval and the runtime settings; use the
config file or environment variables.
Run
enroll,reportandstatusas the service user (sudo -u vultrack-agent …). These commands write the refresh token; run as root, the token file ends up owned by root and the service (uservultrack-agent) can no longer read it.exportdoes not touch the token and can run as root.
Register the agent with the VulTrack server. Requires an enrollment key obtained from the VulTrack web UI.
sudo -u vultrack-agent vultrack-agent enroll --enrollment-key YOUR_ENROLLMENT_KEYOn success, the refresh token is saved to refresh_token_file. The access
token is held in memory only.
To force re-enrollment (revokes the existing registration):
sudo -u vultrack-agent vultrack-agent enroll --enrollment-key YOUR_ENROLLMENT_KEY --forcesudo -u vultrack-agent vultrack-agent reportRuntime facts that need other processes' /proc entries (listeners,
packagesInUse) are reported as insufficient privileges here, because the
capabilities are only granted to the systemd service.
# Recommended: run as a systemd service
sudo systemctl start vultrack-agent
# Or run manually in the foreground (as the service user)
sudo -u vultrack-agent vultrack-agent daemonThe daemon handles SIGINT and SIGTERM for graceful shutdown and sends an
initial report immediately on startup. It automatically refreshes the access
token and re-enrolls using enrollment_key when the refresh token expires.
Collect all data locally and write it as formatted JSON — without contacting the server. Useful for debugging or offline inspection:
sudo vultrack-agent export -o report.json
sudo vultrack-agent export | jq .
sudo vultrack-agent export | jq .runtimeThe export contains the runtime facts (always with the full kernel module index) and does not touch the module index state file. If no usable config file exists, the default runtime settings are used.
Display the active configuration and validate the agent's registration against the server. The check exchanges the refresh token for a new one (the server rotates it on every use), so run it as the service user:
sudo -u vultrack-agent vultrack-agent statusstatus sends no report. If the server rejects the refresh token, the agent
re-enrolls with enrollment_key, exactly as the daemon would.
| Field | Description | Example |
|---|---|---|
hostname |
FQDN of the system | server01.example.com |
osFamily |
OS family | ubuntu, debian, rhel, centos, rocky, alma |
osRelease |
OS version | 24.04, 12, 9.4 |
osCodename |
OS codename (if available) | noble, bookworm |
kernel |
Kernel version (uname -r) |
6.8.0-31-generic |
arch |
System architecture | amd64, arm64 |
ipv4Addrs |
Non-loopback IPv4 addresses | ["192.168.1.10"] |
packageManager |
Detected package manager | dpkg, rpm |
For each installed package:
| Field | Description |
|---|---|
name |
Package name |
version |
Full version string |
arch |
Package architecture |
source |
Source package name |
Packages are collected via dpkg-query on Debian/Ubuntu systems and
rpm -qa on RHEL/CentOS/Rocky/Alma systems.
Every report also carries an optional runtime object. The server uses it to
decide automatically whether a vulnerability can affect the host, for example
"this kernel CVE is in a module that is not loaded and cannot be loaded" or
"this daemon only listens on 127.0.0.1".
Collector (runtime_collectors) |
Field | Content | Source |
|---|---|---|---|
kernel |
kernel |
Kernel release, modulesDisabled, loaded modules (with path), modules blacklisted or disabled in modprobe.d, the kernel module index, selected sysctls, installed kernels, reboot required |
/proc, /lib/modules/<release>, modprobe.d, /var/lib/dpkg/status, /var/run/reboot-required* |
listeners |
listeners |
Listening TCP and bound UDP sockets, also inside containers, with process name, executable, package and systemd unit | /proc/net/*, /proc/<pid>/{fd,comm,exe,cgroup,ns/net} |
packages_in_use |
packagesInUse |
Packages whose files are mapped by running processes, and processes still running a replaced (upgraded) file | /proc/<pid>/{maps,exe,comm} |
package_info |
packageInfo |
Every installed package with the kinds of entry points it ships (executable, service, plugin, hook) and the installed packages it depends on |
/var/lib/dpkg/status, /var/lib/dpkg/info/*.list, lstat of the listed files |
services |
services |
All systemd service units with active, sub and unit file state and package | systemctl list-units, systemctl show |
virtualization |
virtualization |
Virtualization or container type | systemd-detect-virt |
Packages are resolved with the dpkg file lists (/var/lib/dpkg/info/*.list),
including merged /usr and dpkg diversions. On non-dpkg systems the package
fields, installedKernels, rebootRequired, packagesInUse and packageInfo
are omitted.
Processes with their own root filesystem (containers) are reported with
process, executable and unit, but never with a host package, and they do not
count in packagesInUse: their paths belong to the container image, not to the
host's dpkg database. With socket activation (e.g. ssh.socket) the listener
is attributed to the service that inherited the socket, not to systemd.
Package info. The server dismisses findings in libraries that nothing on
the host can use: not loaded by any process for weeks, without programs or
services of their own, and not needed by any package that has some (typically
old library versions left behind after upgrades). packageInfo supplies the
"programs or services of their own" and "needed by" parts:
entryPoints:executablefor files inbin/sbin/libexecdirectories and any file with an execute, setuid or setgid bit;servicefor systemd units and generators, init scripts, cron jobs, D-Bus activation and autostart files;pluginfor shared objects outside the standard library directories, NSS modules and Python/Perl/Ruby/Node.js/PHP modules;hookfor files in directories that make other software run them (profile.d,apt.conf.d,pam.d, udev rules,ld.so.conf.d, …). Paths are compared in their/usrform, also without merged/usr.depends:Depends,Pre-DependsandRecommendswith all alternatives; virtual packages are replaced by their installed providers, names that are not installed are dropped.
Because the server trusts the absence of entry points and dependencies, the
collector over-reports when in doubt: a file it cannot inspect counts as
executable. Packages count as installed in every dpkg state in which their
files may be on disk (installed, half-installed, half-configured,
unpacked, triggers-*), whatever the selection (install, hold, …). The
list is all or nothing: if the dpkg database or a file list cannot be read, or
the time budget runs out, packageInfo is omitted with an error. Only file
names and modes are read, never file contents.
Selected packageInfo entries from an Ubuntu 24.04 server (agent 1.3.0; 836
entries, 88 KB of the 186 KB runtime object):
"packageInfo": [
{"package": "curl", "entryPoints": ["executable"], "depends": ["libc6", "libcurl4t64", "zlib1g"]},
{"package": "libcurl4t64",
"depends": ["ca-certificates", "libbrotli1", "libc6", "libgssapi-krb5-2", "libidn2-0", "libldap2",
"libnghttp2-14", "libpsl5t64", "librtmp1", "libssh-4", "libssl3t64", "libzstd1", "zlib1g"]},
{"package": "libpam-modules", "entryPoints": ["executable", "plugin"],
"depends": ["debconf", "libaudit1", "libc6", "libcrypt1", "libdb5.3t64", "libpam-modules-bin",
"libpam0g", "libselinux1", "libsystemd0"]},
{"package": "libssl3t64", "entryPoints": ["plugin"], "depends": ["libc6"]},
{"package": "openssl", "entryPoints": ["executable"], "depends": ["libc6", "libssl3t64"]}
]libcurl4t64 has no entry points of its own; it stays relevant because curl
depends on it. libssl3t64 is a plugin because of its OpenSSL providers and
engines, libpam-modules because of the PAM modules in security/ (and an
executable helper).
Unknown is not empty. A collector that cannot run (missing permission,
missing tool, error, timeout) is omitted and the reason is listed in
runtime.errors. An empty list means "collected, nothing found". Values that
cannot be determined exactly (e.g. the package of a file) are left out instead
of guessed. Collection problems never prevent the report from being sent.
Not collected: command lines, environment variables, working directories,
open files other than mapped executables and libraries, user names or UIDs of
processes, and the contents of configuration files other than the blacklist
and install directives in modprobe.d.
Privileges. listeners and packages_in_use need to read other
processes' /proc entries. The agent checks for CAP_DAC_READ_SEARCH and
CAP_SYS_PTRACE (root has both; the systemd unit grants them to the service
user). Without them both collectors are reported as insufficient privileges,
because partial data would look complete.
Size and performance. All collectors share the time budget
runtime_timeout_seconds. Lists are capped (2,000 listeners, 3,000 packages
in use, 6,000 packages in packageInfo, 2,000 services, 20 processes per
package); a capped category is listed in runtime.truncated. The runtime object stays below 1 MB; if it would
be larger, the module index lists, then packageInfo, then packagesInUse are
left out (with an entry in errors). The kernel
module index (several thousand lines) is sent in full only when it changed,
when the server has not yet accepted it, or once every 24 hours; otherwise only
its hash is sent. The hash of the accepted index is stored in
module_index_state_file, the only file runtime collection writes.
Example. Abbreviated runtime object from an Ubuntu 24.04 VM running
Docker (agent 1.2.1; container names and IDs anonymised, most list entries
removed). The module index is shown in its hash-only form, as sent when the
server already has it:
{
"schemaVersion": 1,
"collectedAt": "2026-11-05T11:48:58Z",
"kernel": {
"release": "6.8.0-147-generic",
"modulesDisabled": false,
"loadedModules": [
{"name": "nf_tables", "path": "kernel/net/netfilter/nf_tables.ko.zst"},
{"name": "overlay", "path": "kernel/fs/overlayfs/overlay.ko.zst"},
{"name": "tls", "path": "kernel/net/tls/tls.ko.zst"}
],
"blacklistedModules": [
{"name": "algif_aead", "installCommand": "/bin/false"},
{"name": "pcspkr"}
],
"moduleIndex": {"hash": "sha256:a950e3bc590c79f416d2cd92bca51a48c8adec323ad49d585f22dc69411ff160"},
"sysctls": {
"kernel.apparmor_restrict_unprivileged_userns": "1",
"kernel.dmesg_restrict": "1",
"kernel.kptr_restrict": "1",
"kernel.modules_disabled": "0",
"kernel.perf_event_paranoid": "4",
"kernel.unprivileged_bpf_disabled": "2",
"kernel.unprivileged_userns_clone": "1",
"kernel.yama.ptrace_scope": "1",
"user.max_user_namespaces": "47691"
},
"installedKernels": ["6.8.0-137-generic", "6.8.0-147-generic"],
"rebootRequired": false
},
"listeners": [
{"protocol": "tcp", "address": "0.0.0.0", "port": 22, "process": "sshd", "exe": "/usr/sbin/sshd",
"package": "openssh-server", "unit": "ssh.service", "hostNetns": true},
{"protocol": "tcp", "address": "0.0.0.0", "port": 8080, "process": "docker-proxy", "exe": "/usr/bin/docker-proxy",
"package": "docker-ce", "unit": "docker.service", "hostNetns": true},
{"protocol": "tcp", "address": "127.0.0.53", "port": 53, "process": "systemd-resolve", "exe": "/usr/lib/systemd/systemd-resolved",
"package": "systemd-resolved", "unit": "systemd-resolved.service", "hostNetns": true},
{"protocol": "tcp6", "address": "::", "port": 443, "process": "caddy", "exe": "/usr/bin/caddy",
"unit": "docker-4d88e05acf61.scope", "hostNetns": true},
{"protocol": "tcp6", "address": "::", "port": 8080, "process": "app", "exe": "/app/app",
"unit": "docker-20984cf3b843.scope", "hostNetns": false},
{"protocol": "udp", "address": "0.0.0.0", "port": 123, "process": "chronyd", "exe": "/usr/sbin/chronyd",
"package": "chrony", "unit": "chrony.service", "hostNetns": true}
],
"packagesInUse": [
{"package": "docker-ce", "processes": ["docker-proxy", "dockerd"]},
{"package": "libssl3t64", "processes": ["clamd", "freshclam", "fwupd", "sshd", "sudo", "systemd", "systemd-resolve"]},
{"package": "openssh-server", "processes": ["sshd"]}
],
"services": [
{"unit": "docker.service", "activeState": "active", "subState": "running", "unitFileState": "enabled", "package": "docker-ce"},
{"unit": "node_exporter.service", "activeState": "active", "subState": "running", "unitFileState": "enabled"},
{"unit": "ssh.service", "activeState": "active", "subState": "running", "unitFileState": "enabled", "package": "openssh-server"}
],
"virtualization": "kvm",
"truncated": ["packagesInUse"]
}What the example shows:
sshdowns port 22 althoughssh.socket(PID 1) holds the socket as well.caddyruns in a container with host networking (hostNetns: true),appin its own network namespace; neither gets a host package.node_exporterwas installed outside dpkg, so it has nopackage.truncatedlistspackagesInUsebecauselibc6is used by more than 20 distinct processes on this host.
Disabling. Set collect_runtime: false (or VULTRACK_COLLECT_RUNTIME=false)
to omit the runtime facts entirely, or list only the wanted collectors in
runtime_collectors. The server then treats the missing facts as unknown and
never dismisses a finding based on them.
The agent uses two tokens:
| Token | Storage | Lifetime |
|---|---|---|
| Access token (JWT) | Memory only, never written to disk | Short-lived (default 24 h) |
Refresh token (rt_…) |
refresh_token_file (0600) |
Long-lived (default 90 days) |
The agent manages token renewal automatically:
- If a valid access token is in memory — use it directly.
- If the access token is absent or expired — exchange the refresh token for a new pair.
- If the refresh token is rejected (401) — re-enroll automatically using
enrollment_key, unless the agent is waiting for approval (see below). - If the registration is
pending— keep the tokens and wait for admin approval; the agent tries again every cycle and starts reporting once it is approved. - If the registration was revoked by an administrator (403
agent_revoked) — do nothing; the agent does not re-enroll on its own.
Server compatibility. The server signals a pending or revoked registration
with 403 and a code of agent_pending or agent_revoked. Older servers answer
401 instead, which the agent cannot tell apart from an invalid token. With such
a server the agent avoids a re-enrollment loop while it knows it is pending, but
the daemon re-enrolls once after a restart, and a revoked agent re-enrolls (and is
active again at once if its enrollment key auto-approves).
This is why enrollment_key must remain permanently in the config — it enables
fully autonomous recovery without manual intervention.
| Condition | Behavior |
|---|---|
Network error or 5xx |
Retry up to 3 times with exponential backoff (1s, 2s, 4s) |
401 on report |
Refresh access token and retry once |
401 on token refresh |
Re-enroll with force=true using enrollment_key (not while pending approval) |
401 on enrollment |
Enrollment key invalid — log error and exit |
403 with agent_pending |
Waiting for approval — logged as information, retried next cycle, never re-enrolls |
403 with agent_revoked |
Revoked by an administrator — logged as error, never re-enrolls |
other 403 |
Access denied — logged as error, no retry |
409 Conflict on enroll |
Hostname already registered — use --force to re-enroll |
- Refresh token file is stored with
0600permissions (owner read-only) - Refresh token directory is created with
0700permissions - Access token is held in memory only and never written to disk
- Token writes use an atomic write-then-rename pattern to prevent corruption
- TLS certificate validation is enforced by default
- Custom CA certificates can be provided for internal PKI environments
- Tokens and enrollment keys are never written to logs;
statusshows only the first 8 characters of the enrollment key - Ansible tasks use
no_log: truefor steps that handle secrets - The service runs as the unprivileged
vultrack-agentuser with onlyCAP_DAC_READ_SEARCHandCAP_SYS_PTRACE(for runtime facts); the syscallsptrace,process_vm_readv,process_vm_writevandpidfd_getfdare blocked, so the capabilities cannot be used to access other processes' memory
The systemd unit included in the package and in contrib/vultrack-agent.service
provides:
- Automatic restart on failure
- Network dependency (
network-online.target) - Security hardening (
NoNewPrivileges,PrivateTmp, syscall filter, etc.) - The capabilities needed for runtime facts (
CAP_DAC_READ_SEARCH,CAP_SYS_PTRACE) andAF_UNIXforsystemctl - Logging to the systemd journal
sudo journalctl -u vultrack-agent -f
sudo journalctl -u vultrack-agent -n 100
sudo journalctl -u vultrack-agent --since todayvultrack-agent/
├── cmd/
│ └── vultrack-agent/
│ └── main.go # CLI entry point (6 commands)
├── internal/
│ ├── api/
│ │ └── client.go # HTTP client for VulTrack API v2
│ ├── collector/
│ │ ├── system.go # System information collection
│ │ └── packages.go # Package enumeration (dpkg/rpm)
│ ├── runtimefacts/ # Runtime facts (kernel, listeners, packages
│ │ │ # in use, services, virtualization)
│ │ └── testdata/ # Fixture root filesystem and golden file
│ └── config/
│ └── config.go # Configuration loading and validation
├── ansible/
│ └── roles/
│ └── vultrack_agent/ # Ansible role
├── contrib/
│ ├── vultrack-agent.service # Systemd unit file
│ └── config.yaml.example # Annotated example configuration
├── .github/
│ ├── dependabot.yml # Automated dependency updates
│ └── workflows/
│ ├── release.yml # Automated release workflow
│ └── zizmor.yml # GitHub Actions security scanning
├── build.sh # Build script (binary, Debian package; ARCH=amd64 by default)
├── INSTALL.md # Manual installation options
└── README.md
github.com/spf13/cobra— CLI frameworkgo.yaml.in/yaml/v4— YAML parsing- Standard library for everything else
- Go 1.26 or higher
dpkg-debfor creating Debian packages (apt install dpkg-dev)
MIT License