Server Security Init

Server Security Init

Purpose

Use this skill to harden a fresh SSH server without locking the user out. Treat SSH, firewall, and fail2ban changes as high-risk operations that require staged verification.

This skill is intended for Ubuntu/Debian systems using OpenSSH, systemd, UFW, and fail2ban. For other distributions or firewalls, adapt carefully and state the differences before changing anything.

Required Inputs

Collect or infer these before making changes. If any value is ambiguous, ask before proceeding. For a fresh server request, ask for the bootstrap facts first instead of starting remote commands immediately.

| Parameter | Meaning | | --- | --- | | root_public_key_installed | Whether the chosen SSH public key has already been installed for the initial server login user | | target_host | IP address or existing SSH alias for the server | | initial_user | Current login user, often root | | initial_port | Current SSH port, often 22 | | admin_user | Non-root sudo user to create | | public_key_source | Existing remote authorized key, local .pub file, or pasted public key | | local_key_name | Local key basename or server alias to use when selecting or generating a dedicated key | | identity_file | Local private key path for future SSH config | | new_ssh_port | Final SSH port | | ssh_config_alias | Local SSH config alias to create/update | | management_ips | Current and common admin egress IPs or jump hosts to whitelist in fail2ban |

Optional settings:

| Parameter | Default | | --- | --- | | disable_root_login | true | | disable_password_login | true | | enable_ufw | true | | allowed_tcp_ports | only new_ssh_port | | enable_fail2ban | ask or use user preference | | fail2ban_maxretry | 3 | | fail2ban_findtime | 86400 | | fail2ban_bantime | 86400 | | generate_local_keypair | ask during intake if no usable key exists | | local_key_comment | local_key_name or user-specified comment | | local_key_passphrase | ask; do not assume empty passphrase unless user prefers automation |

Input Validation

Validate values before substituting them into shell commands or config files:

• admin_user: must be a normal Linux username such as ^[a-z_][a-z0-9_-]{0,31}$; reject root, empty values, whitespace, shell metacharacters, path separators, and newline characters. If the account already exists, stop unless it is clearly an intended human admin account with UID >= 1000, a real home directory, and an interactive shell; never grant passwordless sudo to service/system accounts.
• initial_port, new_ssh_port, and allowed_tcp_ports: must be numeric TCP ports in 1..65535; prefer unprivileged ports 1024..65535 unless the user explicitly chooses a privileged port.
• public_key_source or public_key_content: must resolve to exactly one valid non-empty SSH public key line. Verify with ssh-keygen -lf against a temporary public-key file and never print private key material.
• target_host, ssh_config_alias, and identity_file: must not contain newline characters. Treat paths and aliases as data, not shell fragments.
• management_ips: must contain only explicit IPs or CIDR ranges that the user recognizes. If the current egress IP cannot be determined confidently, ask before enabling fail2ban.

Safety Rules

• Do not disable root login, password login, the old SSH port, or the old firewall rule until admin_user@new_ssh_port has been verified in a separate SSH command.
• Keep an existing working SSH session open while changing SSH or firewall configuration.
• Back up files before editing: /etc/ssh/sshd_config, relevant /etc/ssh/sshd_config.d/*.conf, fail2ban jail files, and local ~/.ssh/config.
• Prefer drop-in files for new SSH/fail2ban policy, but ensure earlier directives cannot override the intended final state. Verify with sshd -T.
• On systemd hosts, check ssh.socket before and after changing SSH ports. If socket activation is enabled, systemd may own the listening port and ignore Port values in sshd_config; prefer disabling ssh.socket and running the traditional ssh.service unless intentionally managing socket units.
• Use server-side truth for port/firewall verification: ss -ltnp, sshd -T, systemctl status ssh.socket ssh.service, and ufw status. Listener checks must include systemd as well as sshd; socket activation can make PID 1 own the old port. Local tests can be misleading under Clash/Mihomo TUN or fake-IP DNS.
• Configure fail2ban ignoreip before or while enabling fail2ban. Include the current admin egress IP and any usual jump/VPS exits.
• If connectivity is lost, stop changing configuration and recover through an alternate egress, console, rescue panel, or provider VNC before continuing.
• When updating local SSH config, make a backup and replace only the target host block. Do not use broad regexes that can consume following Host blocks.
• Do not bake personal defaults into this skill or its examples. Use placeholders such as `, , , , and `.
• When generating a local keypair, never overwrite an existing private key. If identity_file or identity_file.pub already exists, stop and ask whether to reuse it or choose another name.
• If the user only has an IP and root password, do not use password automation by default. Generate or select a public key, give the user a minimal command to install it under /root/.ssh/authorized_keys, and wait until root public-key login is verified.
• Do not place root passwords in shell commands, SSH config, scripts, logs, or final answers. Password-based bootstrap is an advanced exception only when the user explicitly requests it and the environment has a safe non-interactive method.
• If existing SSH config contains Match blocks or other non-stock policy, stop before bulk-commenting directives and explain the risk. Conditional Match policy can change effective authentication for specific users or source addresses.
• If admin_user already exists, inspect it before writing sudoers or SSH keys. Stop on system users, service users, missing home directories, or non-interactive shells unless the user provides an explicit migration plan.
• If UFW is already active with non-SSH rules, do not reset it without explicit user confirmation and a migration plan for existing service ports.

Workflow

1. Stage -2: intake gate

• When the user asks to initialize or harden a server, ask for these values before touching the server:
• Whether the SSH public key has already been installed for the initial login user.
• Server IP or HostName.
• Initial SSH user and port.
• Local SSH config alias and local key name.
• Non-root passwordless sudo username to create.
• Final SSH port.
• Whether to enable UFW and fail2ban.
• Management IPs or other jump/server IPs to add to fail2ban ignoreip.
• If values are inferable from existing SSH config or prior context, state the inferred values and ask for confirmation before high-risk changes.
• If the public key is not installed yet, do not attempt remote initialization. Move to the key bootstrap gate.

1. Stage -1: key selection or generation

• Use this when the user has no existing key, wants a dedicated per-server key, or has not installed a key on the server yet.
• Read references/ubuntu-openssh-ufw-fail2ban.md for local ssh-keygen command patterns.
• Generate an ed25519 key at the user-specified identity_file only after confirming the key name and path.
• Set public_key_source to the generated .pub file.
• Show the public-key fingerprint, not the private key.
• Use placeholder examples in instructions, for example:

``bash ssh-keygen -t ed25519 -f ~/.ssh/ -C "" ``

1. Stage 0: public-key bootstrap gate

• If root public-key login is not already available, instruct the user to install public_key_source into /root/.ssh/authorized_keys through provider console, Tabby, VNC, or a one-time manual SSH password login.
• Give the user generic server-side commands for the initial login user. For root, use:

``bash install -d -m 700 -o root -g root /root/.ssh touch /root/.ssh/authorized_keys chmod 600 /root/.ssh/authorized_keys chown root:root /root/.ssh/authorized_keys grep -qxF '' /root/.ssh/authorized_keys || printf '%s\n' '' >> /root/.ssh/authorized_keys ``

• If the initial login user is not root, adjust the home directory and ownership for that user.
• Do not proceed to remote initialization until initial_user@target_host:initial_port works with public-key authentication and BatchMode=yes.

1. Preflight

• Confirm reachable public-key login path and gather system facts.
• Inspect effective SSH settings, systemd ssh.socket/ssh.service state, listeners, firewall state, and current authorized keys.
• Determine current admin egress IP from server logs or an external IP service.
• Validate all collected inputs before using them in commands.

1. Stage 1: add the new safe path

• Install sudo and ufw if needed.
• Create admin_user, or verify an existing admin_user is a normal human admin account before changing its keys or sudo privileges.
• Install the selected public key for admin_user.
• Configure passwordless sudo for admin_user and validate with visudo.
• Configure SSH to listen on both initial_port and new_ssh_port.
• Configure UFW to allow both ports temporarily.
• If ssh.socket exists and is active/enabled, disable it, remove or back up socket-related ssh.service drop-ins if present, run systemctl daemon-reload, and enable the traditional SSH service before relying on sshd_config ports.
• Restart SSH and verify the real listener with ss -ltnp including systemd and sshd, then verify admin_user@new_ssh_port plus sudo -n whoami.

1. Stage 2: harden the final state

• Set final SSH policy: Port new_ssh_port, PubkeyAuthentication yes, PermitRootLogin no, PasswordAuthentication no, KbdInteractiveAuthentication no.
• Validate with sshd -t, verify effective policy with sshd -T and sshd -T -C where source/user context matters, ensure ssh.socket cannot re-own the old port after reboot, then restart SSH and re-verify admin_user@new_ssh_port before changing final firewall rules.
• Reset UFW to default deny incoming, default allow outgoing, and allow only required ports.
• Verify with ss -ltnp that SSH listens on new_ssh_port and not the old port, then verify root login is refused and password login is refused.

1. Stage 3: fail2ban

• Read references/ubuntu-openssh-ufw-fail2ban.md before enabling fail2ban.
• Configure the sshd jail with backend = systemd, port = new_ssh_port, and banaction = ufw.
• Include management_ips in ignoreip.
• Restart fail2ban and verify its ban list. If a management IP is banned, unban it immediately and fix ignoreip.

1. Stage 4: local SSH config

• Back up the local SSH config.
• Update or create only ssh_config_alias with the final host, user, port, and identity file.
• Verify with ssh -G ssh_config_alias.

1. Final verification

• Read and follow references/verification-checklist.md.
• Report the exact commands run and the meaningful outputs.

Reference Files

• references/ubuntu-openssh-ufw-fail2ban.md: staged Ubuntu/Debian command patterns and recovery notes.
• references/verification-checklist.md: final test commands and expected results.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: DeerYang
• Source: DeerYang/server-security-init-skill
• License: MIT

Install and usage instructions live in the source repository linked above.